scan2doc_all.py

#!/usr/bin/env python3
"""
Scan2Doc Pro v6.0 — Complete OCR-to-Word Converter
Single file: Engine + GUI + Installer
"""
#!/usr/bin/env python3
"""
Scan2Doc Pro v6.0 — Complete OCR-to-Word Application
======================================================
Complete rewrite with PDF support, speed optimizations, and enhanced processing.

Major improvements over v5:
  - PDF Support: pymupdf (fitz) for PDF-to-image conversion, text & scanned PDFs
  - Speed: bilateralFilter replaces fastNlMeansDenoising, parallel PDF processing
  - Error Handling: try/except blocks, atexit temp cleanup, graceful degradation
  - Enhanced Image Processing: auto-rotate detection, contrast stretching,
    morphological cleanup, adaptive thresholding
  - Better Deskew: Hough line voting for more accurate angle detection
  - All v5 features retained: multi-PSM OCR, cell-by-cell table extraction,
    hOCR parsing, Persian/Arabic support, gridless table detection,
    paragraph reconstruction, heading detection, professional Word output

Supports: Persian (Farsi), Arabic, and English
Engine:   Tesseract OCR + OpenCV + pymupdf + python-docx
"""

import atexit
import argparse
import math
import os
import re
import subprocess
import sys
import tempfile
import time
import warnings
from collections import defaultdict
from concurrent.futures import ThreadPoolExecutor, as_completed
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Callable, Dict, List, Optional, Tuple

import cv2
import numpy as np
import pytesseract
from PIL import Image
from bs4 import BeautifulSoup
from docx import Document
from docx.shared import Inches, Pt, Cm
from docx.enum.text import WD_ALIGN_PARAGRAPH
from docx.enum.table import WD_TABLE_ALIGNMENT
from docx.oxml.ns import qn, nsdecls
from docx.oxml import parse_xml

try:
    import fitz  # pymupdf
    HAS_FITZ = True
except ImportError:
    HAS_FITZ = False
    print("[WARN] pymupdf not installed — PDF support disabled. pip install pymupdf")

warnings.filterwarnings("ignore", category=FutureWarning)

# ═══════════════════════════════════════════════════════════════════════════════
# Temp file cleanup via atexit
# ═══════════════════════════════════════════════════════════════════════════════

_TEMP_FILES: list = []


def _cleanup_temp_files():
    for fp in _TEMP_FILES:
        try:
            if os.path.exists(fp):
                os.unlink(fp)
        except Exception:
            pass
    _TEMP_FILES.clear()


atexit.register(_cleanup_temp_files)


def _register_temp(suffix=".png"):
    """Create a temp file and register for automatic cleanup."""
    fd, path = tempfile.mkstemp(suffix=suffix, prefix="_s2d_v6_")
    os.close(fd)
    _TEMP_FILES.append(path)
    return path


# ═══════════════════════════════════════════════════════════════════════════════
# Constants — Persian/Arabic OCR character fixes (COMPLETE)
# ═══════════════════════════════════════════════════════════════════════════════

PERSIAN_FIXES = {
    # Arabic → Persian letters
    'ك': 'ک', 'ي': 'ی', 'ؤ': 'و', 'إ': 'ا', 'أ': 'ا', 'ٱ': 'ا',
    # Arabic → Persian digits
    '٤': '۴', '٥': '۵', '٦': '۶', '٧': '۷', '٨': '۸', '٩': '۹',
    '٠': '۰', '١': '۱', '٢': '۲', '٣': '۳',
    # Arabic letter forms → Persian
    'ة': 'ه',
    # Remove diacritics (tashkeel)
    'ّ': '', 'ْ': '', 'ٌ': '', 'ً': '', 'ٓ': '', 'ٔ': '',
    # KEEP Persian punctuation (not convert to ASCII!)
    '؟': '؟', '،': '،', '؛': '؛',
    # Remove kashida
    'ـ': '',
    # Unicode normalization
    '\u0640': '',  # Tatweel
    '\u0670': '',  # Superscript alef
    '\u200c': '',  # Zero-width non-joiner
    '\u200d': '',  # Zero-width joiner
    '\u00a0': ' ',  # Non-breaking space
    '\u200b': '',  # Zero-width space
    '\ufeff': '',  # BOM
}


def fix_persian(text: str) -> str:
    """Complete Persian text fix: Arabic→Persian, numbers, diacritics, whitespace."""
    if not text:
        return text
    result = [PERSIAN_FIXES.get(ch, ch) for ch in text]
    text = ''.join(result)
    text = re.sub(r'(?<=[۰-۹])\s*(?=[۰-۹])', '', text)
    text = re.sub(r'\s+', ' ', text)
    return text.strip()


# Alias for backward compatibility
fix_persian_text = fix_persian


def detect_rtl(text: str) -> bool:
    """Detect if text is predominantly RTL (Persian/Arabic/Hebrew)."""
    if not text:
        return False
    rtl_chars = 0
    ltr_chars = 0
    for c in text:
        code = ord(c)
        if (0x0600 <= code <= 0x06FF or 0x0750 <= code <= 0x077F or
            0x0590 <= code <= 0x05FF or 0xFB50 <= code <= 0xFDFF or
            0xFE70 <= code <= 0xFEFF or 0x0590 <= code <= 0x08FF or
            0xFB00 <= code <= 0xFBFF or 0xFE00 <= code <= 0xFEFF):
            rtl_chars += 1
        elif 'a' <= c.lower() <= 'z':
            ltr_chars += 1
    return rtl_chars > ltr_chars


# ═══════════════════════════════════════════════════════════════════════════════
# Data Classes
# ═══════════════════════════════════════════════════════════════════════════════

@dataclass
class BBox:
    x: int; y: int; w: int; h: int

    @property
    def x2(self): return self.x + self.w
    @property
    def y2(self): return self.y + self.h
    @property
    def center_x(self): return self.x + self.w / 2
    @property
    def center_y(self): return self.y + self.h / 2
    @property
    def area(self): return self.w * self.h

    def overlaps(self, other, threshold=0.5):
        x1 = max(self.x, other.x); y1 = max(self.y, other.y)
        x2 = min(self.x2, other.x2); y2 = min(self.y2, other.y2)
        if x2 <= x1 or y2 <= y1: return False
        return (x2 - x1) * (y2 - y1) / max(self.area, 1) > threshold


@dataclass
class TextBlock:
    bbox: BBox
    text: str = ""
    font_size: float = 12.0
    is_bold: bool = False
    is_italic: bool = False
    is_rtl: bool = False
    confidence: float = 0.0
    element_type: str = "paragraph"
    alignment: str = "left"
    column: int = 0


@dataclass
class TableData:
    bbox: BBox
    rows: int = 0
    cols: int = 0
    cells: List[List[str]] = field(default_factory=list)


@dataclass
class LineElement:
    bbox: BBox
    orientation: str = "horizontal"
    is_table_border: bool = False
    thickness: int = 1


@dataclass
class ImageRegion:
    bbox: BBox
    aspect_ratio: float = 0.0


@dataclass
class PageStatistics:
    text_blocks: int = 0
    tables: int = 0
    lines: int = 0
    image_regions: int = 0
    columns: int = 1
    overall_confidence: float = 0.0
    processing_time: float = 0.0
    is_rtl: bool = False
    total_words: int = 0
    low_confidence_words: int = 0


@dataclass
class LayoutResult:
    text_blocks: List[TextBlock] = field(default_factory=list)
    tables: List[TableData] = field(default_factory=list)
    lines: List[LineElement] = field(default_factory=list)
    image_regions: List[ImageRegion] = field(default_factory=list)
    columns: int = 1
    column_boundaries: List[Tuple[int, int]] = field(default_factory=list)
    is_rtl: bool = False
    median_font_size: float = 12.0


# ═══════════════════════════════════════════════════════════════════════════════
# ImageProcessor — v6: PDF support, faster denoise, auto-rotate, enhanced preprocessing
# ═══════════════════════════════════════════════════════════════════════════════

class ImageProcessor:
    def __init__(self, dpi=300):
        self.dpi = dpi

    # ── PDF to Images ──

    def pdf_to_images(self, pdf_path, max_pages=None):
        """Convert PDF pages to images using pymupdf (fitz).
        Returns list of numpy arrays (BGR) for each page."""
        if not HAS_FITZ:
            raise ImportError("pymupdf is required for PDF support: pip install pymupdf")

        try:
            doc = fitz.open(pdf_path)
        except Exception as e:
            print(f"  [PDF] Error opening {pdf_path}: {e}")
            return []

        pages = []
        page_count = len(doc)
        if max_pages:
            page_count = min(page_count, max_pages)

        print(f"  [PDF] {os.path.basename(pdf_path)}: {page_count} page(s)")

        for i in range(page_count):
            try:
                page = doc[i]
                # Render at target DPI
                zoom = self.dpi / 72.0
                mat = fitz.Matrix(zoom, zoom)
                pix = page.get_pixmap(matrix=mat, alpha=False)

                # Convert to numpy array
                img_data = np.frombuffer(pix.samples, dtype=np.uint8)
                img = img_data.reshape(pix.height, pix.width, 3)

                # pymupdf gives RGB, OpenCV expects BGR
                img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
                pages.append(img)
                print(f"  [PDF] Page {i + 1}/{page_count}: {pix.width}x{pix.height}")
            except Exception as e:
                print(f"  [PDF] Warning: Page {i + 1} failed: {e}")
                continue

        doc.close()
        return pages

    def is_scanned_pdf(self, pdf_path):
        """Check if PDF is scanned (image-based) or text-based.
        Returns (is_scanned, text_page_count, image_page_count)."""
        if not HAS_FITZ:
            return True, 0, 0

        try:
            doc = fitz.open(pdf_path)
            text_pages = 0
            image_pages = 0
            for page in doc:
                text = page.get_text().strip()
                if len(text) > 50:
                    text_pages += 1
                else:
                    image_pages += 1
            doc.close()
            is_scanned = image_pages > text_pages
            return is_scanned, text_pages, image_pages
        except Exception:
            return True, 0, 0

    # ── Image Loading ──

    def load_image(self, path):
        """Load image with proper error handling. Supports image files and PDFs."""
        ext = Path(path).suffix.lower()

        if ext == '.pdf':
            pages = self.pdf_to_images(path)
            if not pages:
                raise FileNotFoundError(f"Cannot load PDF: {path}")
            return pages[0]  # Return first page for single-page processing

        img = cv2.imread(path, cv2.IMREAD_COLOR)
        if img is None:
            raise FileNotFoundError(f"Cannot load image: {path}")
        print(f"  [IMG] Loaded: {os.path.basename(path)} ({img.shape[1]}x{img.shape[0]})")
        return img

    # ── Color Handling ──

    def handle_color(self, img):
        """Better color image handling: detect if grayscale vs color and convert appropriately."""
        if len(img.shape) == 2:
            return img  # Already grayscale

        # Check if image is effectively grayscale (low color variance)
        hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
        s_channel = hsv[:, :, 1]
        mean_saturation = np.mean(s_channel)

        if mean_saturation < 15:
            # Low saturation — treat as grayscale for better OCR
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
            print(f"  [IMG] Low saturation ({mean_saturation:.1f}), using grayscale")
            return gray

        # For color images, convert to grayscale but enhance contrast
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        return gray

    # ── Denoising (v6: faster bilateralFilter) ──

    def denoise(self, img, strength=10):
        """Fast denoising using bilateralFilter instead of slow fastNlMeansDenoising.
        bilateralFilter preserves edges while smoothing noise — much faster."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()

        try:
            # bilateralFilter: d=9, sigmaColor=75, sigmaSpace=75
            # Much faster than fastNlMeansDenoising with good edge preservation
            denoised = cv2.bilateralFilter(gray, d=9, sigmaColor=75, sigmaSpace=75)
            return denoised
        except Exception as e:
            print(f"  [IMG] Denoise fallback: {e}")
            return gray

    def denoise_fast(self, img):
        """Ultra-fast denoising for time-critical paths (e.g., parallel processing)."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()
        return cv2.bilateralFilter(gray, d=5, sigmaColor=50, sigmaSpace=50)

    def denoise_aggressive(self, img):
        """Aggressive denoising for very noisy images."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()
        # Step 1: Bilateral filter (edge-preserving)
        filtered = cv2.bilateralFilter(gray, d=9, sigmaColor=75, sigmaSpace=75)
        # Step 2: Morphological opening to remove small noise
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
        opened = cv2.morphologyEx(filtered, cv2.MORPH_OPEN, kernel)
        # Step 3: Median blur for salt-and-pepper noise
        denoised = cv2.medianBlur(opened, 3)
        return denoised

    def detect_noise_level(self, img):
        """Detect noise level in image (0-100)."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()
        # Use Laplacian variance as noise estimate
        laplacian = cv2.Laplacian(gray, cv2.CV_64F)
        noise_level = np.var(laplacian)
        # Normalize to 0-100 scale
        return min(100, noise_level / 100)

    # ── Contrast Enhancement ──

    def enhance_contrast(self, img):
        """Apply contrast stretching for better OCR preprocessing."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()

        # Check if image is already good contrast
        p_low, p_high = np.percentile(gray, [2, 98])
        if p_high - p_low > 100:
            return gray  # Already good contrast

        # Multi-stage contrast enhancement
        # Stage 1: CLAHE for local contrast
        clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))
        enhanced = clahe.apply(gray)

        # Stage 2: Contrast stretching
        p_low, p_high = np.percentile(enhanced, [1, 99])
        if p_high - p_low > 10:
            enhanced = np.clip((enhanced.astype(np.float32) - p_low) / (p_high - p_low) * 255, 0, 255)
            enhanced = enhanced.astype(np.uint8)

        # Stage 3: Sharpening for very blurry images
        variance = np.var(cv2.Laplacian(enhanced, cv2.CV_64F))
        if variance < 100:  # Very blurry
            kernel = np.array([[-1,-1,-1], [-1,9,-1], [-1,-1,-1]])
            enhanced = cv2.filter2D(enhanced, -1, kernel)
            enhanced = np.clip(enhanced, 0, 255).astype(np.uint8)

        return enhanced

    # ── Auto-rotate Detection ──

    def detect_rotation(self, img):
        """Detect text orientation and suggest rotation angle.
        Uses minAreaRect on text pixels for dominant orientation."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()

        # Use adaptive thresholding for better text detection
        try:
            binary = cv2.adaptiveThreshold(
                gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                cv2.THRESH_BINARY_INV, 31, 10)
        except Exception:
            _, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)

        # Dilate to connect text into regions
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
        dilated = cv2.dilate(binary, kernel, iterations=2)

        coords = np.column_stack(np.where(dilated > 0))
        if len(coords) < 100:
            return 0.0

        # Use minAreaRect to find dominant orientation
        angle = cv2.minAreaRect(coords)[-1]
        if angle < -45:
            angle = 90 + angle

        # Only return significant angles
        if abs(angle) < 0.5:
            return 0.0

        # Clamp to reasonable range
        return max(-15.0, min(15.0, angle))

    # ── Improved Deskew ──

    def deskew(self, img):
        """Improved deskew using Hough line voting for more accurate angle detection.
        Falls back to minAreaRect if Hough detection fails."""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()

        # Method 1: Hough line voting (more accurate for document images)
        edges = cv2.Canny(gray, 50, 150, apertureSize=3)
        lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 100,
                                minLineLength=gray.shape[1] // 4, maxLineGap=10)

        if lines is not None and len(lines) > 5:
            angles = []
            for line in lines:
                pts = line.reshape(4)
                x1, y1, x2, y2 = int(pts[0]), int(pts[1]), int(pts[2]), int(pts[3])
                if abs(x2 - x1) > abs(y2 - y1):  # Mostly horizontal lines
                    angle = math.degrees(math.atan2(y2 - y1, x2 - x1))
                    if abs(angle) < 45:  # Consider wider range of angles
                        angles.append(angle)

            if angles:
                # Use median angle (robust to outliers)
                angle = np.median(angles)
                if abs(angle) < 0.3:
                    return img, 0.0
                h, w = img.shape[:2]
                M = cv2.getRotationMatrix2D((w // 2, h // 2), angle, 1.0)
                # Calculate new bounding box to avoid cropping
                cos_a = abs(M[0, 0])
                sin_a = abs(M[0, 1])
                new_w = int(h * sin_a + w * cos_a)
                new_h = int(h * cos_a + w * sin_a)
                M[0, 2] += (new_w - w) / 2
                M[1, 2] += (new_h - h) / 2
                rotated = cv2.warpAffine(img, M, (new_w, new_h), flags=cv2.INTER_CUBIC,
                                         borderMode=cv2.BORDER_REPLICATE)
                print(f"  [IMG] Deskewed by {angle:.2f}° (Hough)")
                return rotated, angle

        # Method 2: Fallback to minAreaRect
        inverted = cv2.bitwise_not(gray)
        coords = np.column_stack(np.where(inverted > 0))
        if len(coords) < 50:
            return img, 0.0
        angle = cv2.minAreaRect(coords)[-1]
        if angle < -45:
            angle = 90 + angle
        if abs(angle) < 0.3:
            return img, 0.0
        h, w = img.shape[:2]
        M = cv2.getRotationMatrix2D((w // 2, h // 2), angle, 1.0)
        rotated = cv2.warpAffine(img, M, (w, h), flags=cv2.INTER_CUBIC,
                                 borderMode=cv2.BORDER_REPLICATE)
        print(f"  [IMG] Deskewed by {angle:.2f}° (minAreaRect)")
        return rotated, angle

    # ── Binarization (v6: multiple methods) ──

    def binarize(self, img, method="adaptive"):
        """Binarize image using specified method.
        Methods: 'adaptive' (default), 'otsu', 'sauvola'"""
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()

        if method == "otsu":
            _, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
            return binary
        elif method == "sauvola":
            # Sauvola-like adaptive threshold (simulated with Gaussian adaptive)
            return cv2.adaptiveThreshold(
                gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                cv2.THRESH_BINARY, 31, 12)
        else:  # adaptive
            return cv2.adaptiveThreshold(
                gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                cv2.THRESH_BINARY, 31, 10)

    # ── Morphological Cleanup ──

    def morphological_cleanup(self, binary):
        """Clean up binarized image with morphological operations.
        Remove small noise dots and fill small holes."""
        if len(binary.shape) == 3:
            gray = cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
        else:
            gray = binary.copy()

        # Remove small noise (opening)
        kernel_small = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
        cleaned = cv2.morphologyEx(gray, cv2.MORPH_OPEN, kernel_small)

        # Fill small holes (closing)
        kernel_medium = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
        cleaned = cv2.morphologyEx(cleaned, cv2.MORPH_CLOSE, kernel_medium)

        return cleaned

    # ── Main Preparation Pipeline ──

    def prepare_for_ocr(self, path):
        """Full preprocessing pipeline for file input."""
        img = self.load_image(path)
        gray = self.handle_color(img)
        enhanced = self.enhance_contrast(gray)
        # Smart denoising based on noise level
        noise_level = self.detect_noise_level(enhanced)
        if noise_level > 50:
            print(f"  [IMG] High noise detected ({noise_level:.0f}), using aggressive denoising")
            denoised = self.denoise_aggressive(enhanced)
        else:
            denoised = self.denoise(enhanced)
        deskewed, angle = self.deskew(denoised)
        binary = self.binarize(deskewed)
        binary = self.morphological_cleanup(binary)
        return deskewed, binary

    def prepare_for_ocr_from_array(self, img):
        """Preprocessing pipeline for numpy array input (e.g., PDF pages).
        Returns (gray, binary) for downstream processing."""
        gray = self.handle_color(img)
        enhanced = self.enhance_contrast(gray)
        # Smart denoising based on noise level
        noise_level = self.detect_noise_level(enhanced)
        if noise_level > 50:
            denoised = self.denoise_aggressive(enhanced)
        else:
            denoised = self.denoise(enhanced)
        deskewed, angle = self.deskew(denoised)
        binary = self.binarize(deskewed)
        binary = self.morphological_cleanup(binary)
        return deskewed, binary


# ═══════════════════════════════════════════════════════════════════════════════
# LayoutAnalyzer — v6: Enhanced with better error handling
# ═══════════════════════════════════════════════════════════════════════════════

class LayoutAnalyzer:
    def __init__(self, min_block_area=100):
        self.min_block_area = min_block_area

    # ── Multi-Scale Line Detection ──

    def detect_lines_multiscale(self, binary, min_length_ratio=0.10):
        """Detect horizontal and vertical lines at multiple scales.
        Enhanced with error handling for edge cases."""
        h, w = binary.shape
        if h < 10 or w < 10:
            return [], []

        inverted = ~binary if len(binary.shape) == 2 else ~cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
        all_h, all_v = [], []

        for scale in [8, 10, 15, 20, 30, 40]:
            try:
                # Horizontal lines
                kw = max(w // scale, 8)
                kh_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (kw, 1))
                h_mask = cv2.morphologyEx(inverted, cv2.MORPH_OPEN, kh_kernel)
                contours, _ = cv2.findContours(h_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
                for c in contours:
                    x, y, cw, ch = cv2.boundingRect(c)
                    if cw > w * min_length_ratio:
                        all_h.append(LineElement(BBox(x, y, cw, max(ch, 2)), "horizontal", thickness=max(ch, 1)))

                # Vertical lines
                kv = cv2.getStructuringElement(cv2.MORPH_RECT, (1, max(h // scale, 8)))
                v_mask = cv2.morphologyEx(inverted, cv2.MORPH_OPEN, kv)
                contours, _ = cv2.findContours(v_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
                for c in contours:
                    x, y, cw, ch = cv2.boundingRect(c)
                    if ch > h * min_length_ratio:
                        all_v.append(LineElement(BBox(x, y, max(cw, 2), ch), "vertical", thickness=max(cw, 1)))
            except Exception as e:
                continue  # Skip failed scale gracefully

        # Hough lines for additional precision
        try:
            gray = binary if len(binary.shape) == 2 else cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
            edges = cv2.Canny(gray, 50, 150, apertureSize=3)
            min_len = max(w * 0.20, 80)
            lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 150,
                                    minLineLength=min_len, maxLineGap=10)
            if lines is not None:
                for line in lines:
                    x1, y1, x2, y2 = line[:4]
                    dx, dy = abs(x2 - x1), abs(y2 - y1)
                    length = math.sqrt(dx * dx + dy * dy)
                    if length < min_len * 0.4:
                        continue
                    if dy < 5 and dx > dy * 3:
                        all_h.append(LineElement(BBox(min(x1, x2), min(y1, y2),
                                                     abs(x2 - x1), max(abs(y2 - y1), 2)),
                                                 "horizontal", thickness=2))
                    elif dx < 5 and dy > dx * 3:
                        all_v.append(LineElement(BBox(min(x1, x2), min(y1, y2),
                                                     max(abs(x2 - x1), 2), abs(y2 - y1)),
                                                 "vertical", thickness=2))
        except Exception:
            pass

        # Detect underlines
        self._detect_underlines(inverted, w, h, all_h)

        all_h = self._merge_lines(all_h, "horizontal")
        all_v = self._merge_lines(all_v, "vertical")
        return all_h, all_v

    def _detect_underlines(self, inverted, w, h, all_lines):
        """Detect short horizontal underlines (typically under headings)."""
        for kw_ratio in [4, 5, 6]:
            try:
                kw = max(w // kw_ratio, 30)
                kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (kw, 1))
                mask = cv2.morphologyEx(inverted, cv2.MORPH_OPEN, kernel)
                contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
                for c in contours:
                    x, y, cw, ch = cv2.boundingRect(c)
                    if w * 0.05 < cw < w * 0.40 and ch < h * 0.005:
                        all_lines.append(LineElement(BBox(x, y, cw, max(ch, 2)),
                                                    "horizontal", thickness=max(ch, 1)))
            except Exception:
                continue

    def _merge_lines(self, lines, orientation):
        if not lines:
            return []
        key = (lambda l: l.bbox.y) if orientation == "horizontal" else (lambda l: l.bbox.x)
        lines.sort(key=key)
        merged = []
        for line in lines:
            if merged:
                last = merged[-1]
                close = abs(key(line) - key(last)) < 8
                if orientation == "horizontal":
                    close = close and abs(line.bbox.center_x - last.bbox.center_x) < last.bbox.w * 0.5
                else:
                    close = close and abs(line.bbox.center_y - last.bbox.center_y) < last.bbox.h * 0.5
                if close:
                    x1 = min(last.bbox.x, line.bbox.x)
                    y1 = min(last.bbox.y, line.bbox.y)
                    x2 = max(last.bbox.x2, line.bbox.x2)
                    y2 = max(last.bbox.y2, line.bbox.y2)
                    merged[-1] = LineElement(BBox(x1, y1, x2 - x1, y2 - y1), orientation,
                                             thickness=max(last.thickness, line.thickness))
                    continue
            merged.append(line)
        return merged

    # ── Table Detection ──

    def detect_tables(self, binary, h_lines, v_lines):
        h, w = binary.shape
        tables = []

        long_h = [l for l in h_lines if l.bbox.w > w * 0.15]
        long_v = [l for l in v_lines if l.bbox.h > h * 0.10]

        if len(long_h) < 2 or len(long_v) < 1:
            return tables

        combined = np.zeros((h, w), dtype=np.uint8)
        for line in long_h:
            cv2.line(combined, (line.bbox.x, line.bbox.y), (line.bbox.x2, line.bbox.y2),
                     255, max(line.thickness, 1))
        for line in long_v:
            cv2.line(combined, (line.bbox.x, line.bbox.y), (line.bbox.x2, line.bbox.y2),
                     255, max(line.thickness, 1))

        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
        dilated = cv2.dilate(combined, kernel, iterations=2)

        num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(dilated, 8)

        for i in range(1, num_labels):
            x, y = stats[i, cv2.CC_STAT_LEFT], stats[i, cv2.CC_STAT_TOP]
            bw, bh = stats[i, cv2.CC_STAT_WIDTH], stats[i, cv2.CC_STAT_HEIGHT]
            area = stats[i, cv2.CC_STAT_AREA]
            if bw < w * 0.10 or bh < h * 0.02 or area < 400:
                continue

            h_in = sum(1 for l in long_h if l.bbox.y >= y - 5 and l.bbox.y <= y + bh + 5)
            v_in = sum(1 for l in long_v if l.bbox.x >= x - 5 and l.bbox.x <= x + bw + 5)
            if h_in < 2 or v_in < 1:
                continue

            t_hl = [l for l in long_h if l.bbox.y >= y - 5 and l.bbox.y <= y + bh + 5]
            t_vl = [l for l in long_v if l.bbox.x >= x - 5 and l.bbox.x <= x + bw + 5]

            table = TableData(bbox=BBox(x, y, bw, bh))
            table.rows = max(len(t_hl) - 1, 1)
            table.cols = max(len(t_vl) - 1, 1)
            tables.append(table)
            print(f"  [LAYOUT] Table: {x},{y} {bw}x{bh} ({table.rows}x{table.cols})")

        if len(tables) > 1:
            tables.sort(key=lambda t: t.bbox.area, reverse=True)
            kept = []
            for t in tables:
                if not any(t.bbox.overlaps(e.bbox, 0.3) for e in kept):
                    kept.append(t)
            tables = kept

        return tables

    # ── Gridless Table Detection ──

    def detect_gridless_tables(self, binary, text_blocks, existing_tables):
        """Detect tables WITHOUT grid lines — just aligned text in rows/columns."""
        h, w = binary.shape
        if not text_blocks or len(text_blocks) < 4:
            return []

        existing_bboxes = [t.bbox for t in existing_tables]

        sorted_blocks = sorted(text_blocks, key=lambda b: (b.bbox.y, -b.bbox.x if b.is_rtl else b.bbox.x))

        rows = []
        current_row = [sorted_blocks[0]]
        for block in sorted_blocks[1:]:
            prev = current_row[-1]
            prev_cy = prev.bbox.center_y
            block_cy = block.bbox.center_y
            row_height = max(prev.bbox.h, block.bbox.h)
            if abs(prev_cy - block_cy) < row_height * 0.6:
                current_row.append(block)
            else:
                rows.append(current_row)
                current_row = [block]
        rows.append(current_row)

        if len(rows) < 3:
            return []

        tables = []
        i = 0
        while i < len(rows) - 2:
            group = [rows[i]]
            row_xs = [b.bbox.x for b in sorted(rows[i], key=lambda b: b.bbox.x)]

            for j in range(i + 1, min(i + 8, len(rows))):
                next_xs = [b.bbox.x for b in sorted(rows[j], key=lambda b: b.bbox.x)]
                if len(next_xs) >= 2 and len(row_xs) >= 2:
                    if len(next_xs) == len(row_xs):
                        mismatch = sum(1 for a, b in zip(row_xs, next_xs)
                                     if abs(a - b) > w * 0.10)
                        if mismatch <= len(row_xs) * 0.3:
                            group.append(rows[j])
                elif len(next_xs) == len(row_xs):
                    group.append(rows[j])

            if len(group) >= 3:
                all_blocks = [b for row in group for b in row]
                min_x = min(b.bbox.x for b in all_blocks)
                min_y = min(b.bbox.y for b in all_blocks)
                max_x2 = max(b.bbox.x2 for b in all_blocks)
                max_y2 = max(b.bbox.y2 for b in all_blocks)
                bbox = BBox(min_x, min_y, max_x2 - min_x, max_y2 - min_y)

                if any(bbox.overlaps(t, 0.3) for t in existing_bboxes):
                    i += len(group)
                    continue

                num_cols = max(len(row) for row in group)
                table = TableData(bbox=bbox, rows=len(group), cols=num_cols)
                tables.append(table)
                print(f"  [LAYOUT] Gridless table: {bbox.x},{bbox.y} ({len(group)}x{num_cols})")
                i += len(group)
            else:
                i += 1

        return tables

    # ── Text Block Detection ──

    def detect_text_blocks(self, binary):
        """Detect text blocks with enhanced splitting based on line gaps."""
        h, w = binary.shape
        if h < 10 or w < 10:
            return []

        try:
            tk = cv2.getStructuringElement(cv2.MORPH_RECT, (30, 5))
            if len(binary.shape) == 3:
                gray_bin = cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
            else:
                gray_bin = binary
            tm = cv2.morphologyEx(~gray_bin, cv2.MORPH_CLOSE, tk)
            tm = cv2.dilate(tm, cv2.getStructuringElement(cv2.MORPH_RECT, (20, 10)), iterations=2)
        except Exception as e:
            print(f"  [LAYOUT] Block detection error: {e}")
            return []

        blocks = []
        for c in cv2.findContours(tm, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]:
            x, y, bw, bh = cv2.boundingRect(c)
            if bw * bh < self.min_block_area or bw < w * 0.03 or bh < 8:
                continue
            blocks.append(BBox(x, y, bw, bh))

        # Split large blocks based on internal line gaps
        split_blocks = []
        for block in blocks:
            if block.h < 30:
                split_blocks.append(block)
                continue

            try:
                region = gray_bin[block.y:block.y2, block.x:block.x2]
                inverted = ~region
                h_proj = np.sum(inverted, axis=1).astype(np.float64)

                kernel_size = max(3, int(block.h * 0.03))
                if kernel_size % 2 == 0: kernel_size += 1
                if kernel_size > 1:
                    h_proj_s = cv2.GaussianBlur(h_proj.reshape(1, -1), (1, kernel_size), 0).flatten()
                else:
                    h_proj_s = h_proj

                threshold = np.max(h_proj_s) * 0.05 if np.max(h_proj_s) > 0 else 0
                gap_start = None
                gaps = []
                for y_pos in range(len(h_proj_s)):
                    if h_proj_s[y_pos] <= threshold:
                        if gap_start is None:
                            gap_start = y_pos
                    else:
                        if gap_start is not None:
                            gap_len = y_pos - gap_start
                            if gap_len > block.h * 0.15:
                                gaps.append((gap_start, y_pos))
                            gap_start = None

                if gaps and len(gaps) >= 1:
                    prev_y = 0
                    for gap_start, gap_end in gaps:
                        if gap_start - prev_y > 8:
                            split_blocks.append(BBox(
                                block.x, block.y + prev_y,
                                block.w, gap_start - prev_y))
                        prev_y = gap_end
                    if block.y2 - (block.y + prev_y) > 8:
                        split_blocks.append(BBox(
                            block.x, block.y + prev_y,
                            block.w, block.y2 - (block.y + prev_y)))
                else:
                    split_blocks.append(block)
            except Exception:
                split_blocks.append(block)

        split_blocks.sort(key=lambda b: (b.y, b.x))
        return split_blocks

    # ── Column Detection ──

    def detect_columns(self, binary, text_blocks=None):
        """Detect columns using vertical projection + horizontal word distribution."""
        h, w = binary.shape
        if h < 10 or w < 10:
            return 1, [(0, w)]

        if len(binary.shape) == 3:
            gray_bin = cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
        else:
            gray_bin = binary

        inverted = ~gray_bin
        h_proj = np.sum(inverted, axis=0).astype(np.float64)
        ks = max(w // 100, 5)
        if ks % 2 == 0: ks += 1
        h_proj_s = cv2.GaussianBlur(h_proj.reshape(1, -1), (ks, 1), 0).flatten()

        threshold = np.max(h_proj_s) * 0.02
        in_text = h_proj_s > threshold
        gaps = []
        gs = 0
        in_gap = False
        for x in range(w):
            if not in_text[x]:
                if not in_gap: gs = x; in_gap = True
            else:
                if in_gap and (x - gs) > w * 0.05:
                    gaps.append((gs, x))
                in_gap = False
        if in_gap and (w - gs) > w * 0.05:
            gaps.append((gs, w))

        valid = []
        margin = w * 0.08
        for gs_, ge in gaps:
            if gs_ < margin or ge > w - margin:
                continue
            gap_col = inverted[:, gs_:ge]
            density = np.sum(gap_col > 0) / (h * max(ge - gs_, 1))
            if density < 0.03:
                valid.append((gs_, ge))

        merged = []
        for g in valid:
            if merged and g[0] - merged[-1][1] < w * 0.02:
                merged[-1] = (merged[-1][0], g[1])
            else:
                merged.append(g)

        num_cols = 1 + len(merged)
        boundaries = []
        if merged:
            boundaries.append((0, merged[0][0]))
            for i in range(len(merged) - 1):
                boundaries.append((merged[i][1], merged[i + 1][0]))
            boundaries.append((merged[-1][1], w))
        else:
            boundaries = [(0, w)]

        # Horizontal word distribution analysis
        if num_cols == 1 and text_blocks and len(text_blocks) > 3:
            num_cols, boundaries = self._analyze_word_distribution(
                gray_bin, text_blocks, w, h, boundaries)

        if num_cols > 1:
            print(f"  [LAYOUT] {num_cols} columns: {[(b[0], b[1]) for b in boundaries]}")
        return num_cols, boundaries

    def _analyze_word_distribution(self, binary, text_blocks, w, h, current_boundaries):
        """Analyze horizontal word distribution to detect columns."""
        text_mask = np.zeros((h, w), dtype=np.uint8)
        for block in text_blocks:
            bx, by = max(0, block.x if hasattr(block, "x") else block.bbox.x), \
                     max(0, block.y if hasattr(block, "y") else block.bbox.y)
            bx2, by2 = min(w, block.x2 if hasattr(block, "x2") else block.bbox.x2), \
                        min(h, block.y2 if hasattr(block, "y2") else block.bbox.y2)
            text_mask[by:by2, bx:bx2] = 1

        v_proj = np.sum(text_mask, axis=0).astype(np.float64)
        if np.max(v_proj) == 0:
            return 1, current_boundaries

        ks = max(w // 60, 7)
        if ks % 2 == 0: ks += 1
        v_proj_smooth = cv2.GaussianBlur(v_proj.reshape(1, -1), (ks, 1), 0).flatten()
        v_proj_norm = v_proj_smooth / np.max(v_proj_smooth)

        gap_threshold = 0.05
        in_gap = False
        gap_start = 0
        found_gaps = []
        for x in range(w):
            if v_proj_norm[x] < gap_threshold:
                if not in_gap:
                    gap_start = x
                    in_gap = True
            else:
                if in_gap:
                    gap_width = x - gap_start
                    if gap_width > w * 0.03:
                        gap_region = text_mask[:, gap_start:x]
                        vertical_coverage = np.sum(np.any(gap_region > 0, axis=1)) / h
                        if vertical_coverage < 0.3:
                            found_gaps.append((gap_start, x))
                    in_gap = False

        if found_gaps:
            merged_gaps = [found_gaps[0]]
            for g in found_gaps[1:]:
                if g[0] - merged_gaps[-1][1] < w * 0.01:
                    merged_gaps[-1] = (merged_gaps[-1][0], g[1])
                else:
                    merged_gaps.append(g)

            if len(merged_gaps) >= 1:
                num_cols = len(merged_gaps) + 1
                boundaries = [(0, merged_gaps[0][0])]
                for i in range(len(merged_gaps) - 1):
                    boundaries.append((merged_gaps[i][1], merged_gaps[i + 1][0]))
                boundaries.append((merged_gaps[-1][1], w))
                return num_cols, boundaries

        return 1, current_boundaries

    # ── Image Region Detection ──

    def detect_image_regions(self, binary):
        h, w = binary.shape
        if h < 10 or w < 10:
            return []

        if len(binary.shape) == 3:
            gray = cv2.cvtColor(binary, cv2.COLOR_BGR2GRAY)
        else:
            gray = binary.copy()

        try:
            gx = cv2.Sobel(gray, cv2.CV_64F, 1, 0, ksize=3)
            gy = cv2.Sobel(gray, cv2.CV_64F, 0, 1, ksize=3)
            gm = np.sqrt(gx ** 2 + gy ** 2)
            gt = np.percentile(gm, 80)
            hg = (gm > gt).astype(np.uint8) * 255
            hg = cv2.dilate(hg, cv2.getStructuringElement(cv2.MORPH_RECT, (15, 15)), iterations=2)
        except Exception:
            return []

        regions = []
        for c in cv2.findContours(hg, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]:
            x, y, bw, bh = cv2.boundingRect(c)
            if bw * bh < 2000 or bw < w * 0.05 or bh < h * 0.03:
                continue
            regions.append(ImageRegion(BBox(x, y, bw, bh), bw / max(bh, 1)))
        return regions

    # ── Main Pipeline ──

    def analyze(self, binary, original):
        result = LayoutResult()
        h, w = binary.shape

        h_lines, v_lines = self.detect_lines_multiscale(binary)
        result.lines = h_lines + v_lines

        tables = self.detect_tables(binary, h_lines, v_lines)
        result.tables = tables

        for t in tables:
            for l in result.lines:
                if t.bbox.overlaps(l.bbox, 0.1):
                    l.is_table_border = True

        tb = self.detect_text_blocks(binary)
        tb = [b for b in tb if not any(b.overlaps(t.bbox, 0.3) for t in tables)]

        img_regions = self.detect_image_regions(binary)
        result.image_regions = img_regions
        tb = [b for b in tb if not any(b.overlaps(ir.bbox, 0.3) for ir in img_regions)]

        nc, bnd = self.detect_columns(binary, tb)
        result.columns = nc
        result.column_boundaries = bnd

        # Gridless table detection
        tb_wrapped = [TextBlock(bbox=b) if isinstance(b, BBox) else b for b in tb]
        gridless = self.detect_gridless_tables(binary, tb_wrapped, tables)
        result.tables.extend(gridless)

        for b in tb:
            result.text_blocks.append(TextBlock(bbox=b))

        result.is_rtl = self._detect_rtl(original)
        print(f"  [LAYOUT] {len(result.text_blocks)} blocks, {len(result.tables)} tables, "
              f"{len(result.lines)} lines, {len(result.image_regions)} images, {nc} cols")
        return result

    def _detect_rtl(self, img):
        if len(img.shape) == 3:
            gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        else:
            gray = img.copy()
        _, b = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
        h, w = b.shape
        return np.sum(b[:, w // 2:] > 0) > np.sum(b[:, :w // 2] > 0) * 1.1


# ═══════════════════════════════════════════════════════════════════════════════
# TableExtractor — v6: Enhanced with error handling and contrast options
# ═══════════════════════════════════════════════════════════════════════════════

class TableExtractor:
    def __init__(self, ocr_lang="fas+ara+eng"):
        self.ocr_lang = ocr_lang

    def extract_table(self, gray, table, ocr_engine):
        """Extract table content with cell-by-cell OCR, upscaling, and CLAHE."""
        h, w = gray.shape
        x1 = max(0, table.bbox.x); y1 = max(0, table.bbox.y)
        x2 = min(w, table.bbox.x2); y2 = min(h, table.bbox.y2)
        table_img = gray[y1:y2, x1:x2]
        if table_img.size == 0:
            return TableData(bbox=table.bbox)

        try:
            _, binary = cv2.threshold(table_img, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
        except Exception:
            return TableData(bbox=table.bbox)

        th, tw = table_img.shape

        hp = self._find_positions(binary, axis=1, length=tw)
        vp = self._find_positions(binary, axis=0, length=th)

        if len(hp) < 2 or len(vp) < 2:
            text = ocr_engine.ocr_region(table_img, self.ocr_lang)
            return TableData(bbox=table.bbox, rows=1, cols=1, cells=[[fix_persian_text(text)]])

        rows, cols = len(hp) - 1, len(vp) - 1
        if rows > 50 or cols > 20:
            text = ocr_engine.ocr_region(table_img, self.ocr_lang)
            return TableData(bbox=table.bbox, rows=1, cols=1, cells=[[fix_persian_text(text)]])

        cells = []
        for r in range(rows):
            rc = []
            for c in range(cols):
                cy1, cy2 = hp[r] + 3, hp[r + 1] - 3
                cx1, cx2 = vp[c] + 3, vp[c + 1] - 3
                if cy2 <= cy1 or cx2 <= cx1:
                    rc.append(""); continue
                cy1, cy2 = max(0, cy1), min(th, cy2)
                cx1, cx2 = max(0, cx1), min(tw, cx2)
                cell = table_img[cy1:cy2, cx1:cx2]
                if cell.size == 0:
                    rc.append(""); continue

                # Skip empty cells
                dark_pixels = np.sum(cell < 128)
                if dark_pixels < 10:
                    rc.append(""); continue

                try:
                    # Upscale cell (3x)
                    ch, cw = cell.shape
                    upscale_factor = 3
                    upscaled = cv2.resize(cell, (cw * upscale_factor, ch * upscale_factor),
                                         interpolation=cv2.INTER_CUBIC)

                    # Apply CLAHE
                    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(4, 4))
                    enhanced = clahe.apply(upscaled)

                    # Add padding
                    pad = 15
                    eh, ew = enhanced.shape
                    padded = np.ones((eh + pad * 2, ew + pad * 2), dtype=np.uint8) * 255
                    padded[pad:pad + eh, pad:pad + ew] = enhanced

                    text = ocr_engine.ocr_region_best(padded, self.ocr_lang)
                    rc.append(fix_persian_text(text).strip())
                except Exception as e:
                    rc.append("")
            cells.append(rc)

        non_empty_rows = [row for row in cells if any(c.strip() for c in row)]
        if not non_empty_rows:
            non_empty_rows = cells

        actual_rows = len(non_empty_rows)
        print(f"  [TABLE] Extracted {actual_rows}x{cols} table (from {rows}x{cols} grid)")
        return TableData(bbox=table.bbox, rows=actual_rows, cols=cols, cells=non_empty_rows)

    def _find_positions(self, binary, axis, length):
        """Find line positions using morphological operations at multiple scales."""
        positions = set()
        for scale in [3, 4, 5, 6]:
            try:
                k = max(length // scale, 10)
                if axis == 1:
                    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (k, 1))
                    mask = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel)
                    proj = np.sum(mask, axis=1)
                else:
                    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, k))
                    mask = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel)
                    proj = np.sum(mask, axis=0)

                thresh = np.max(proj) * 0.3
                in_line = proj > thresh
                start = None
                for i, val in enumerate(in_line):
                    if val:
                        if start is None: start = i
                    else:
                        if start is not None:
                            positions.add((start + i) // 2)
                            start = None
                if start is not None:
                    positions.add((start + len(in_line) - 1) // 2)
            except Exception:
                continue

        if not positions:
            return []
        sp = sorted(positions)
        clusters = [[sp[0]]]
        for p in sp[1:]:
            if p - clusters[-1][-1] < 5:
                clusters[-1].append(p)
            else:
                clusters.append([p])
        result = [int(np.mean(cl)) for cl in clusters]

        shape_len = binary.shape[0] if axis == 1 else binary.shape[1]
        if len(result) >= 2:
            avg_col_width = (result[-1] - result[0]) / (len(result) - 1)
            if result[0] > avg_col_width * 0.8:
                result.insert(0, 0)
            if shape_len - result[-1] > avg_col_width * 0.8:
                result.append(shape_len - 1)
        else:
            if not result or result[0] > shape_len * 0.15:
                result.insert(0, 0)
            if not result or result[-1] < shape_len * 0.85:
                result.append(shape_len - 1)
        return result


# ═══════════════════════════════════════════════════════════════════════════════
# OCREngine — v6: Multi-PSM with confidence-based selection + better error handling
# ═══════════════════════════════════════════════════════════════════════════════

class OCREngine:
    def __init__(self, lang="fas+ara+eng", dpi=300):
        self.lang = lang
        self.dpi = dpi

    def ocr_region(self, img, lang=None):
        """Simple OCR with PSM 6."""
        lang = lang or self.lang
        try:
            if img.size == 0 or img.shape[0] < 3 or img.shape[1] < 3:
                return ""
            return pytesseract.image_to_string(Image.fromarray(img), lang=lang,
                                               config="--psm 6").strip()
        except Exception as e:
            print(f"  [OCR] Warning: {e}")
            return ""

    def ocr_region_best(self, img, lang=None):
        """Multi-PSM OCR with confidence-based selection — picks best result."""
        lang = lang or self.lang
        if img.size == 0 or img.shape[0] < 3 or img.shape[1] < 3:
            return ""

        pil = Image.fromarray(img)
        h, w = img.shape[:2]

        if h < 50 and w > h * 3:
            psms = [7, 13, 8]
        elif h < 80:
            psms = [7, 8, 6]
        else:
            psms = [3, 4, 6, 11]

        best_text, best_conf = "", -1

        for psm in psms:
            try:
                data = pytesseract.image_to_data(pil, lang=lang,
                    config=f'--psm {psm} --oem 3',
                    output_type=pytesseract.Output.DICT)
                confs = [int(c) for c in data['conf'] if int(c) > 0]
                avg_conf = sum(confs) / len(confs) if confs else 0

                text = pytesseract.image_to_string(pil, lang=lang,
                    config=f'--psm {psm} --oem 3').strip()

                if avg_conf > best_conf:
                    best_conf = avg_conf
                    best_text = text
            except Exception:
                continue

        return best_text

    def ocr_full_page(self, img):
        """Full page OCR with PSM 3."""
        try:
            return pytesseract.image_to_string(Image.fromarray(img), lang=self.lang,
                                               config="--psm 3").strip()
        except Exception as e:
            print(f"  [OCR] Warning: {e}")
            return ""

    def ocr_with_hocr(self, img):
        """Generate hOCR via tesseract subprocess with 2x upscaling."""
        try:
            if len(img.shape) == 2:
                work_img = cv2.resize(img, (img.shape[1]*2, img.shape[0]*2),
                                     interpolation=cv2.INTER_CUBIC)
            else:
                work_img = cv2.resize(img, (img.shape[1]*2, img.shape[0]*2),
                                     interpolation=cv2.INTER_CUBIC)

            img_path = _register_temp(".png")
            cv2.imwrite(img_path, work_img)

            effective_dpi = self.dpi * 2
            result = subprocess.run(
                ['tesseract', img_path, 'stdout', '--dpi', str(effective_dpi),
                 '--psm', '3', '-l', self.lang, 'hocr'],
                capture_output=True, text=True, timeout=60
            )
            # Cleanup immediately (atexit will catch stragglers)
            try:
                os.unlink(img_path)
                if img_path in _TEMP_FILES:
                    _TEMP_FILES.remove(img_path)
            except Exception:
                pass
            return result.stdout
        except subprocess.TimeoutExpired:
            print("  [OCR] Warning: hOCR generation timed out")
            return ""
        except Exception as e:
            print(f"  [OCR] Warning: hOCR generation failed: {e}")
            return ""

    def parse_hocr_blocks(self, hocr_text):
        """Parse hOCR into structured blocks with font size, bold, italic, confidence."""
        blocks = []
        if not hocr_text:
            return blocks
        try:
            soup = BeautifulSoup(hocr_text, "html.parser")
        except Exception:
            try:
                soup = BeautifulSoup(hocr_text, "xml")
            except Exception:
                return blocks

        paras = soup.find_all("p", class_="ocr_par")
        if not paras:
            paras = soup.find_all("span", class_="ocr_par")

        for par in paras:
            info = {"bbox": None, "text": "", "font_size": 12.0,
                    "bold": False, "italic": False, "is_rtl": False,
                    "confidence": 85.0, "alignment": "left", "words": []}

            info["bbox"] = self._parse_bbox(par)

            words = par.find_all("span", class_="ocrx_word")
            texts, sizes, confs, bolds = [], [], [], []
            for w in words:
                t = w.get_text(strip=True)
                if not t:
                    continue
                t = fix_persian_text(t)
                texts.append(t)
                fs = self._parse_font_size(w)
                conf = self._parse_confidence(w)
                info["words"].append({
                    "text": t, "bbox": self._parse_bbox(w),
                    "font_size": fs, "bold": self._is_bold(w),
                    "italic": self._is_italic(w), "confidence": conf})
                if fs > 0: sizes.append(fs)
                if conf > 0: confs.append(conf)
                bolds.append(self._is_bold(w))

            info["text"] = " ".join(texts)
            if sizes:
                info["font_size"] = max(set(sizes), key=sizes.count)
            if confs:
                info["confidence"] = sum(confs) / len(confs)
            info["bold"] = any(bolds)
            info["is_rtl"] = detect_rtl(info["text"])

            if info["bbox"] and info["words"]:
                x_positions = [w.get("bbox", {}).get("x", 0) for w in info["words"] if w.get("bbox")]
                if x_positions:
                    avg_x = sum(x_positions) / len(x_positions)
                    bx = info["bbox"].get("x", 0)
                    bw = info["bbox"].get("w", 1)
                    if abs(avg_x - bx) < bw * 0.15:
                        info["alignment"] = "center"

            if info["text"]:
                blocks.append(info)

        return blocks

    def _parse_bbox(self, el):
        title = el.get("title", "")
        if not title: return None
        m = re.search(r'bbox\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)', title)
        if m:
            return {"x": int(m.group(1)), "y": int(m.group(2)),
                    "w": int(m.group(3)) - int(m.group(1)),
                    "h": int(m.group(4)) - int(m.group(2))}
        return None

    def _parse_font_size(self, el):
        title = el.get("title", "")
        m = re.search(r'x_fsize\s+([\d.]+)', title)
        if m: return float(m.group(1))
        m = re.search(r'x_size\s+([\d.]+)', title)
        if m: return float(m.group(1))
        bb = self._parse_bbox(el)
        if bb and bb["h"] > 0: return bb["h"] / 1.2
        return 12.0

    def _parse_confidence(self, el):
        title = el.get("title", "")
        m = re.search(r'x_wconf\s+(\d+)', title)
        if m: return float(m.group(1))
        return 85.0

    def _is_bold(self, el):
        title = str(el.get("title", "")).lower()
        return "bold" in title

    def _is_italic(self, el):
        title = str(el.get("title", "")).lower()
        return "italic" in title

    def get_overall_confidence(self, hocr_text):
        confs = []
        try:
            soup = BeautifulSoup(hocr_text, "html.parser")
        except Exception:
            return 0.0
        for w in soup.find_all("span", class_="ocrx_word"):
            m = re.search(r'x_wconf\s+(\d+)', w.get("title", ""))
            if m: confs.append(int(m.group(1)))
        return sum(confs) / len(confs) if confs else 0.0

    def count_low_confidence_words(self, hocr_text, threshold=70):
        count = 0
        try:
            soup = BeautifulSoup(hocr_text, "html.parser")
        except Exception:
            return 0
        for w in soup.find_all("span", class_="ocrx_word"):
            m = re.search(r'x_wconf\s+(\d+)', w.get("title", ""))
            if m and int(m.group(1)) < threshold:
                count += 1
        return count


# ═══════════════════════════════════════════════════════════════════════════════
# ParagraphReconstructor — v6: Enhanced with error handling
# ═══════════════════════════════════════════════════════════════════════════════

class ParagraphReconstructor:
    def __init__(self, median_font_size=12.0):
        self.median_font_size = median_font_size

    def reconstruct(self, blocks, column_boundaries=None):
        if not blocks:
            return []
        blocks = [b for b in blocks if b.get("text", "").strip()]
        if not blocks:
            return []

        is_rtl = detect_rtl(" ".join(b.get("text", "") for b in blocks))
        if is_rtl:
            blocks.sort(key=lambda b: (b.get("bbox", {}).get("y", 0),
                                       -(b.get("bbox", {}).get("x", 0))))
        else:
            blocks.sort(key=lambda b: (b.get("bbox", {}).get("y", 0),
                                       b.get("bbox", {}).get("x", 0)))

        paragraphs = []
        current = blocks[0]
        for block in blocks[1:]:
            if self._should_merge(current, block):
                current = self._merge(current, block)
            else:
                paragraphs.append(current)
                current = block
        paragraphs.append(current)

        result = []
        for para in paragraphs:
            bd = para.get("bbox")
            if not bd: continue
            try:
                tb = TextBlock(
                    bbox=BBox(**bd), text=para.get("text", "").strip(),
                    font_size=para.get("font_size", 12.0),
                    is_bold=para.get("bold", False),
                    is_italic=para.get("italic", False),
                    is_rtl=para.get("is_rtl", False),
                    confidence=para.get("confidence", 85.0),
                    alignment=para.get("alignment", "left"))
                result.append(tb)
            except Exception as e:
                print(f"  [RECON] Warning: Failed to create TextBlock: {e}")
        return result

    def _should_merge(self, b1, b2):
        bb1, bb2 = b1.get("bbox"), b2.get("bbox")
        if not bb1 or not bb2: return False
        gap = bb2["y"] - (bb1["y"] + bb1["h"])
        max_gap = max(b1.get("font_size", 12) * 1.5, 15)
        if gap > max_gap or gap < -bb1["h"] * 0.5: return False
        xd = abs(bb1["x"] - bb2["x"])
        if xd > max(bb1["w"], bb2["w"], 1) * 0.3: return False
        fs1, fs2 = b1.get("font_size", 12), b2.get("font_size", 12)
        if fs1 > 0 and fs2 > 0 and min(fs1, fs2) / max(fs1, fs2) < 0.7: return False
        if bb2["x"] > bb1["x"] + bb1["w"] * 1.2: return False
        return True

    def _merge(self, b1, b2):
        bb1, bb2 = b1.get("bbox", {}), b2.get("bbox", {})
        x1 = min(bb1.get("x", 0), bb2.get("x", 0))
        y1 = min(bb1.get("y", 0), bb2.get("y", 0))
        x2 = max(bb1.get("x", 0) + bb1.get("w", 0), bb2.get("x", 0) + bb2.get("w", 0))
        y2 = max(bb1.get("y", 0) + bb1.get("h", 0), bb2.get("y", 0) + bb2.get("h", 0))
        return {
            "bbox": {"x": x1, "y": y1, "w": x2 - x1, "h": y2 - y1},
            "text": b1.get("text", "") + " " + b2.get("text", ""),
            "font_size": max(b1.get("font_size", 12), b2.get("font_size", 12)),
            "bold": b1.get("bold", False) or b2.get("bold", False),
            "italic": b1.get("italic", False) or b2.get("italic", False),
            "is_rtl": b1.get("is_rtl", False) or b2.get("is_rtl", False),
            "confidence": min(b1.get("confidence", 85), b2.get("confidence", 85)),
            "words": b1.get("words", []) + b2.get("words", []),
            "alignment": b1.get("alignment", "left")}


# ═══════════════════════════════════════════════════════════════════════════════
# HeadingDetector — v6: Enhanced multi-factor classification
# ═══════════════════════════════════════════════════════════════════════════════

class HeadingDetector:
    def __init__(self, median_font_size=12.0):
        self.median_font_size = median_font_size

    def detect_font_sizes(self, paragraphs):
        sizes = [p.font_size for p in paragraphs if p.font_size > 0]
        if not sizes: return 12.0
        ss = sorted(sizes)
        return ss[len(ss) // 2]

    def classify_paragraphs(self, paragraphs, page_width=2000):
        if not paragraphs: return paragraphs
        self.median_font_size = self.detect_font_sizes(paragraphs)

        bold_count = sum(1 for p in paragraphs if p.is_bold)
        center_count = sum(1 for p in paragraphs if p.alignment == "center")

        for p in paragraphs:
            p.element_type = self._classify(p, page_width, bold_count, center_count)
        return paragraphs

    def _classify(self, para, page_width, bold_count, center_count):
        fs = para.font_size
        text = para.text.strip()
        if not text: return "paragraph"

        if fs > self.median_font_size * 1.8:
            if para.alignment == "center" or len(text) < 100:
                return "title"
        if fs > self.median_font_size * 2.0:
            return "title"

        if fs > self.median_font_size * 1.4:
            return "heading"
        if fs > self.median_font_size * 1.2 and len(text) < 100:
            return "heading"

        if para.is_bold and len(text) < 80 and fs > self.median_font_size * 1.05:
            return "heading"

        if para.alignment == "center" and len(text) < 100 and fs > self.median_font_size * 1.05:
            return "subtitle"

        if para.is_bold and para.alignment == "center" and len(text) < 50:
            return "subtitle"

        if para.is_bold and para.bbox.y < page_width * 0.05 and len(text) < 60:
            return "heading"

        return "paragraph"


# ═══════════════════════════════════════════════════════════════════════════════
# DocumentBuilder — v6: Professional Word document with enhanced styles
# ═══════════════════════════════════════════════════════════════════════════════

class DocumentBuilder:
    def __init__(self, title=None, font_name="B Nazanin"):
        self.title = title
        self.font_name = font_name
        self.doc = None
        self._setup()

    def _setup(self):
        self.doc = Document()
        sec = self.doc.sections[0]
        sec.page_width = Cm(21.0); sec.page_height = Cm(29.7)
        sec.top_margin = Cm(2.5); sec.bottom_margin = Cm(2.5)
        sec.left_margin = Cm(2.5); sec.right_margin = Cm(2.5)

        style = self.doc.styles['Normal']
        style.font.name = self.font_name; style.font.size = Pt(12)
        style.paragraph_format.space_after = Pt(6)
        style.paragraph_format.line_spacing = Pt(18)

        for lvl, sz in [('Heading 1', 18), ('Heading 2', 15), ('Heading 3', 13)]:
            try:
                s = self.doc.styles[lvl]
                s.font.name = self.font_name; s.font.size = Pt(sz); s.font.bold = True
                s.paragraph_format.space_before = Pt(12); s.paragraph_format.space_after = Pt(6)
            except Exception:
                pass

        footer = sec.footer
        footer.is_linked_to_previous = False
        p = footer.paragraphs[0] if footer.paragraphs else footer.add_paragraph()
        p.alignment = WD_ALIGN_PARAGRAPH.CENTER
        r = p.add_run()
        r._r.append(parse_xml(f'<w:fldChar {nsdecls("w")} w:fldCharType="begin"/>'))
        r2 = p.add_run()
        r2._r.append(parse_xml(f'<w:instrText {nsdecls("w")} xml:space="preserve"> PAGE </w:instrText>'))
        r3 = p.add_run()
        r3._r.append(parse_xml(f'<w:fldChar {nsdecls("w")} w:fldCharType="end"/>'))

        if self.title:
            h = self.doc.add_heading(self.title, level=0)
            h.alignment = WD_ALIGN_PARAGRAPH.CENTER
            for r in h.runs:
                r.font.name = self.font_name; r.font.size = Pt(22)

    def _rtl(self, p):
        p._p.get_or_add_pPr().set(qn('w:rtl'), '1')
    def _ltr(self, p):
        p._p.get_or_add_pPr().set(qn('w:ltr'), '1')

    def add_text_block(self, block):
        text = block.text.strip()
        if not text: return
        if block.element_type == "title":
            p = self.doc.add_heading(level=0); p.alignment = WD_ALIGN_PARAGRAPH.CENTER
        elif block.element_type == "heading":
            p = self.doc.add_heading(level=1)
        elif block.element_type == "subtitle":
            p = self.doc.add_heading(level=2); p.alignment = WD_ALIGN_PARAGRAPH.CENTER
        else:
            p = self.doc.add_paragraph()

        for r in p.runs: r.clear()
        r = p.add_run(text)
        r.font.name = self.font_name
        r.font.size = Pt(max(min(block.font_size, 72), 8))
        if block.is_bold: r.bold = True
        if block.is_italic: r.italic = True

        if block.is_rtl:
            self._rtl(p); p.alignment = WD_ALIGN_PARAGRAPH.RIGHT
        else:
            self._ltr(p)
            if block.alignment == "center": p.alignment = WD_ALIGN_PARAGRAPH.CENTER

        pf = p.paragraph_format
        if block.element_type in ("title", "heading"):
            pf.space_before = Pt(12); pf.space_after = Pt(6)
        else:
            pf.space_before = Pt(3); pf.space_after = Pt(3)
            pf.line_spacing = Pt(block.font_size * 1.5)

    def add_table(self, td):
        if td.rows == 0 or td.cols == 0: return
        has = any(c.strip() for row in td.cells for c in row)
        if not has: return

        table = self.doc.add_table(rows=td.rows, cols=td.cols)
        table.style = 'Table Grid'
        table.alignment = WD_TABLE_ALIGNMENT.CENTER
        for r in range(td.rows):
            for c in range(td.cols):
                cell = table.cell(r, c); cell.text = ""
                tc = cell._tc
                tcPr = tc.get_or_add_tcPr()
                tcMar = parse_xml(
                    f'<w:tcMar {nsdecls("w")}>'
                    f'  <w:top w:w="50" w:type="dxa"/>'
                    f'  <w:left w:w="100" w:type="dxa"/>'
                    f'  <w:bottom w:w="50" w:type="dxa"/>'
                    f'  <w:right w:w="100" w:type="dxa"/>'
                    f'</w:tcMar>')
                tcPr.append(tcMar)

                if r < len(td.cells) and c < len(td.cells[r]):
                    t = td.cells[r][c]
                    if t:
                        pp = cell.paragraphs[0]
                        rr = pp.add_run(t)
                        rr.font.name = self.font_name; rr.font.size = Pt(10)
                        if detect_rtl(t):
                            self._rtl(pp); pp.alignment = WD_ALIGN_PARAGRAPH.RIGHT
                        else:
                            self._ltr(pp)
        p = self.doc.add_paragraph()
        p.paragraph_format.space_before = Pt(3); p.paragraph_format.space_after = Pt(3)

    def add_line(self, line):
        if line.orientation == "horizontal":
            p = self.doc.add_paragraph()
            thick = min(max(line.thickness * 2, 4), 12)
            p._p.get_or_add_pPr().append(parse_xml(
                f'<w:pBdr {nsdecls("w")}><w:bottom w:val="single" w:sz="{thick}" w:space="1" w:color="000000"/></w:pBdr>'))
            p.paragraph_format.space_before = Pt(2); p.paragraph_format.space_after = Pt(2)

    def add_image_region(self, image_path, bbox, page_width=2000):
        if image_path and os.path.exists(image_path):
            try:
                w_in = min(bbox.w / 96.0, 6.0)
                self.doc.add_picture(image_path, width=Inches(w_in))
                self.doc.paragraphs[-1].alignment = WD_ALIGN_PARAGRAPH.CENTER
            except Exception as e:
                print(f"  [DOC] Warning: image insert failed: {e}")
        else:
            p = self.doc.add_paragraph("[تصویر]")
            p.alignment = WD_ALIGN_PARAGRAPH.CENTER

    def add_page_break(self):
        self.doc.add_page_break()

    def save(self, output_path):
        self.doc.save(output_path)
        mb = os.path.getsize(output_path) / (1024 * 1024)
        print(f"  [DOC] Saved: {output_path} ({mb:.2f} MB)")


# ═══════════════════════════════════════════════════════════════════════════════
# QualityScorer — v6: Enhanced confidence-based quality scoring
# ═══════════════════════════════════════════════════════════════════════════════

class QualityScorer:
    """Calculate quality score from 0 to 10 based on page statistics."""

    @staticmethod
    def calculate(stats):
        """
        Score formula (v6 enhanced):
          Base: 5.0
          +1.0 if tables found
          +1.5 if confidence > 90, +1.0 if > 80, +0.5 if > 70
          +0.5 if blocks > 2, +0.3 if blocks > 0
          +1.0 if low_confidence_words == 0, +0.5 if < 3
          +0.5 if lines found
          +0.5 if columns > 1
          +0.5 if word count > 0 (text was found)
          Max: 10.0
        """
        score = 5.0

        if stats.get('tables_found', 0) > 0:
            score += 1.0

        conf = stats.get('confidence', 0)
        if conf > 90:
            score += 1.5
        elif conf > 80:
            score += 1.0
        elif conf > 70:
            score += 0.5

        blocks = stats.get('blocks', 0)
        if blocks > 2:
            score += 0.5
        elif blocks > 0:
            score += 0.3

        low_conf = stats.get('low_conf_words', 0)
        if low_conf == 0:
            score += 1.0
        elif low_conf < 3:
            score += 0.5

        if stats.get('lines_found', 0) > 0:
            score += 0.5

        if stats.get('columns', 1) > 1:
            score += 0.5

        if stats.get('word_count', 0) > 0:
            score += 0.5

        return min(score, 10.0)


# ═══════════════════════════════════════════════════════════════════════════════
# BatchProcessor — v6: PDF support, parallel processing, enhanced error handling
# ═══════════════════════════════════════════════════════════════════════════════

class BatchProcessor:
    SUPPORTED = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif', '.webp', '.gif', '.pdf'}

    def __init__(self, lang="fas+ara+eng", dpi=300, title=None,
                 font_name="B Nazanin", progress_callback=None,
                 max_workers=None):
        self.lang = lang
        self.dpi = dpi
        self.title = title
        self.font_name = font_name
        self.progress_callback = progress_callback
        self.max_workers = max_workers or min(4, os.cpu_count() or 1)
        self.img_proc = ImageProcessor(dpi=dpi)
        self.ocr_engine = OCREngine(lang=lang, dpi=dpi)
        self.layout_analyzer = LayoutAnalyzer()
        self.table_extractor = TableExtractor(ocr_lang=lang)
        self.paragraph_reconstructor = ParagraphReconstructor()
        self.heading_detector = HeadingDetector()
        self.quality_scorer = QualityScorer()
        self.stats = []

    def collect_images(self, inputs):
        """Collect images and PDFs from input paths.
        For PDFs, each page becomes a separate entry."""
        images = []
        for inp in inputs:
            p = Path(inp)
            if p.is_dir():
                for ext in self.SUPPORTED:
                    images.extend(str(x) for x in p.glob(f"*{ext}"))
                    images.extend(str(x) for x in p.glob(f"*{ext.upper()}"))
            elif p.is_file() and p.suffix.lower() in self.SUPPORTED:
                if p.suffix.lower() == '.pdf':
                    # For PDFs, expand to individual page entries
                    pages = self._expand_pdf(str(p))
                    images.extend(pages)
                else:
                    images.append(str(p))
            else:
                print(f"  [WARN] Skipping unsupported file: {inp}")

        # Sort: keep page-annotated entries grouped by original PDF
        images.sort(key=lambda x: (x.split('::')[0], int(x.split('::')[1]) if '::' in x else 0))
        return list(dict.fromkeys(images))

    def _expand_pdf(self, pdf_path):
        """Expand a PDF into per-page entries for batch processing.
        Returns list of 'pdf_path::page_index' strings."""
        if not HAS_FITZ:
            print(f"  [WARN] PDF support disabled (pymupdf not installed): {pdf_path}")
            return []

        try:
            doc = fitz.open(pdf_path)
            count = len(doc)
            doc.close()
            return [f"{pdf_path}::{i}" for i in range(count)]
        except Exception as e:
            print(f"  [WARN] Cannot read PDF {pdf_path}: {e}")
            return []

    def _progress(self, cur, total, msg):
        if self.progress_callback:
            self.progress_callback(cur, total, msg)
        else:
            print(f"  [{cur}/{total}] {msg}")

    def process_single_page(self, image_path, page_num, total):
        """Process a single page (image file or PDF page)."""
        ps = PageStatistics()
        try:
            print(f"\n{'=' * 60}")
            print(f"  Page {page_num}/{total}: {os.path.basename(image_path)}")
            print(f"{'=' * 60}")
            t0 = time.time()

            # Handle PDF pages (encoded as 'pdf_path::page_index')
            if '::' in image_path and image_path.endswith(('::0', '::1', '::2', '::3',
                                                            '::4', '::5', '::6', '::7',
                                                            '::8', '::9')) is False:
                pass  # Regular image
            elif '::' in image_path:
                pdf_path, page_idx = image_path.rsplit('::', 1)
                page_idx = int(page_idx)
                gray, binary = self._process_pdf_page(pdf_path, page_idx)
            else:
                gray, binary = self.img_proc.prepare_for_ocr(image_path)

            layout = self.layout_analyzer.analyze(binary, gray)

            hocr = self.ocr_engine.ocr_with_hocr(gray)
            hocr_blocks = self.ocr_engine.parse_hocr_blocks(hocr)
            print(f"  [OCR] Found {len(hocr_blocks)} hOCR blocks")

            paragraphs = self.paragraph_reconstructor.reconstruct(
                hocr_blocks, layout.column_boundaries)
            paragraphs = self.heading_detector.classify_paragraphs(
                paragraphs, page_width=gray.shape[1])
            layout.text_blocks = paragraphs

            # Extract tables with cell-by-cell OCR
            extracted = []
            for t in layout.tables:
                try:
                    extracted.append(self.table_extractor.extract_table(gray, t, self.ocr_engine))
                except Exception as e:
                    print(f"  [TABLE] Warning: Table extraction failed: {e}")
            layout.tables = extracted

            oc = self.ocr_engine.get_overall_confidence(hocr)
            lwc = self.ocr_engine.count_low_confidence_words(hocr)

            self._build_page(layout, gray)

            ps.text_blocks = len(layout.text_blocks)
            ps.tables = len(layout.tables)
            ps.lines = len([l for l in layout.lines if not l.is_table_border])
            ps.image_regions = len(layout.image_regions)
            ps.columns = layout.columns
            ps.overall_confidence = oc
            ps.is_rtl = layout.is_rtl
            ps.total_words = sum(len(b.text.split()) for b in layout.text_blocks)
            ps.low_confidence_words = lwc
            ps.processing_time = time.time() - t0

            print(f"  [STATS] {ps.text_blocks} blocks, {ps.tables} tables, "
                  f"{ps.lines} lines, conf={oc:.0f}%, {ps.processing_time:.1f}s")
            if lwc: print(f"  [WARN] {lwc} low-confidence words")

            page_stats = {
                'tables_found': ps.tables,
                'confidence': oc,
                'blocks': ps.text_blocks,
                'low_conf_words': lwc,
                'lines_found': ps.lines,
                'columns': ps.columns,
                'word_count': ps.total_words,
            }
            page_quality = self.quality_scorer.calculate(page_stats)
            print(f"  [QUALITY] Page score: {page_quality}/10")

            self._progress(page_num, total, f"{ps.text_blocks} blocks, {ps.tables} tables, quality={page_quality}/10")
            return ps

        except Exception as e:
            print(f"  [ERROR] Page {page_num}: {e}")
            import traceback; traceback.print_exc()
            return ps

    def _process_pdf_page(self, pdf_path, page_idx):
        """Convert a single PDF page to grayscale and binary images."""
        if not HAS_FITZ:
            raise ImportError("pymupdf required for PDF support")

        doc = fitz.open(pdf_path)
        page = doc[page_idx]
        zoom = self.dpi / 72.0
        mat = fitz.Matrix(zoom, zoom)
        pix = page.get_pixmap(matrix=mat, alpha=False)

        img_data = np.frombuffer(pix.samples, dtype=np.uint8)
        img = img_data.reshape(pix.height, pix.width, 3)
        img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
        doc.close()

        print(f"  [PDF] Page {page_idx + 1}: {pix.width}x{pix.height}")
        return self.img_proc.prepare_for_ocr_from_array(img)

    def _build_page(self, layout, gray):
        elements = []
        for b in layout.text_blocks:
            if b.text.strip():
                elements.append(("text", b.bbox.y, b))
        for t in layout.tables:
            elements.append(("table", t.bbox.y, t))
        for l in layout.lines:
            if not l.is_table_border:
                elements.append(("line", l.bbox.y, l))
        for ir in layout.image_regions:
            elements.append(("image", ir.bbox.y, ir))

        elements.sort(key=lambda e: e[1])
        for et, yp, elem in elements:
            try:
                if et == "text": self.doc_builder.add_text_block(elem)
                elif et == "table": self.doc_builder.add_table(elem)
                elif et == "line" and elem.orientation == "horizontal":
                    self.doc_builder.add_line(elem)
                elif et == "image":
                    ip = self._extract_img(gray, elem)
                    self.doc_builder.add_image_region(ip, elem.bbox, gray.shape[1])
            except Exception as e:
                print(f"  [DOC] Warning: Failed to add element: {e}")

    def _extract_img(self, gray, ir):
        try:
            h, w = gray.shape
            x1, y1 = max(0, ir.bbox.x), max(0, ir.bbox.y)
            x2, y2 = min(w, ir.bbox.x2), min(h, ir.bbox.y2)
            if x2 <= x1 or y2 <= y1: return None
            region = gray[y1:y2, x1:x2]
            tmp_path = _register_temp(".png")
            cv2.imwrite(tmp_path, region)
            return tmp_path
        except Exception:
            return None

    def process_batch(self, input_paths, output_path, title=None):
        """Process multiple images/PDFs into a single Word document."""
        images = self.collect_images(input_paths)
        if not images:
            print("[ERROR] No valid images or PDFs!")
            return False
        total = len(images)
        print(f"\n{'#' * 60}")
        print(f"  Scan2Doc Pro v6.0 — OCR-to-Word with PDF Support")
        print(f"  {total} page(s), langs={self.lang}, output={output_path}")
        if not HAS_FITZ:
            print(f"  [WARN] pymupdf not installed — PDF support disabled")
        print(f"{'#' * 60}")

        self.doc_builder = DocumentBuilder(title=title, font_name=self.font_name)
        ok, fail = 0, []
        t0 = time.time()

        for i, ip in enumerate(images, 1):
            try:
                ps = self.process_single_page(ip, i, total)
                if ps.text_blocks > 0 or ps.tables > 0:
                    ok += 1; self.stats.append(ps)
                else:
                    fail.append(i)
            except Exception as e:
                print(f"  [ERROR] Page {i}: {e}")
                import traceback; traceback.print_exc()
                fail.append(i)
            if i < total: self.doc_builder.add_page_break()

        try:
            self.doc_builder.save(output_path)
        except Exception as e:
            print(f"[ERROR] Save failed: {e}")
            return False

        tt = time.time() - t0
        print(f"\n{'#' * 60}")
        print(f"  Done! {ok}/{total} pages, {tt:.1f}s")
        if fail: print(f"  Failed: {fail}")

        if self.stats:
            ac = sum(s.overall_confidence for s in self.stats) / len(self.stats)
            tb = sum(s.text_blocks for s in self.stats)
            tt2 = sum(s.tables for s in self.stats)
            tl = sum(s.lines for s in self.stats)
            tw = sum(s.total_words for s in self.stats)
            lc = sum(s.low_confidence_words for s in self.stats)

            print(f"\n  === AGGREGATE STATISTICS ===")
            print(f"  Blocks: {tb}")
            print(f"  Tables: {tt2}")
            print(f"  Lines: {tl}")
            print(f"  Words: {tw}")
            print(f"  Confidence: {ac:.1f}%")
            print(f"  Low-confidence words: {lc}")

            agg_stats = {
                'tables_found': tt2,
                'confidence': ac,
                'blocks': tb,
                'low_conf_words': lc,
                'lines_found': tl,
                'columns': max((s.columns for s in self.stats), default=1),
                'word_count': tw,
            }
            quality = self.quality_scorer.calculate(agg_stats)
            print(f"\n  === QUALITY SCORE: {quality}/10 ===")
        print(f"{'#' * 60}")
        return ok > 0


# ═══════════════════════════════════════════════════════════════════════════════
# CLI — v6: Same interface as v5, with PDF support
# ═══════════════════════════════════════════════════════════════════════════════

def build_parser():
    p = argparse.ArgumentParser(
        prog="scan2doc_pro_v6",
        description="Scan2Doc Pro v6.0 — OCR-to-Word Converter with PDF Support")
    p.add_argument("-i", "--input", nargs="+", required=True,
                   help="Input image(s) or PDF(s)")
    p.add_argument("-o", "--output", required=True, help="Output .docx path")
    p.add_argument("--lang", default="fas+ara+eng", help="Tesseract language(s)")
    p.add_argument("--dpi", type=int, default=300, help="Scan DPI")
    p.add_argument("--title", default=None, help="Document title")
    p.add_argument("--font", default="B Nazanin", help="Font name")
    p.add_argument("--workers", type=int, default=None,
                   help="Max parallel workers (default: auto)")
    return p


def main():
    args = build_parser().parse_args()
    if not args.output.lower().endswith(".docx"):
        args.output += ".docx"
    proc = BatchProcessor(lang=args.lang, dpi=args.dpi, title=args.title,
                          font_name=args.font, max_workers=args.workers)
    ok = proc.process_batch(args.input, args.output, title=args.title)
    sys.exit(0 if ok else 1)


if __name__ == "__main__":
    main()

# ═══════════════════════════════════════════════════════════════
# GUI Section
# ═══════════════════════════════════════════════════════════════

#!/usr/bin/env python3
"""
Scan2Doc Pro — Professional GUI Application
============================================
Modern CustomTkinter-based desktop application for OCR-to-Word conversion.
Features: image preview, real-time logging, batch processing, quality scoring.
"""

import os
import sys
import threading
import time
import traceback
from pathlib import Path

# ── Headless-safe import for customtkinter ──
os.environ.setdefault("DISPLAY", ":0")

try:
    import customtkinter as ctk
except ImportError:
    print("ERROR: customtkinter not installed. Run: pip install customtkinter")
    sys.exit(1)

# Set theme
ctk.set_appearance_mode("dark")
ctk.set_default_color_theme("blue")

try:
    from PIL import Image, ImageTk
    PIL_AVAILABLE = True
except ImportError:
    PIL_AVAILABLE = False

# Import OCR engine
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
try:
    from scan2doc_pro_v6 import BatchProcessor, QualityScorer, PageStatistics
except ImportError:
    # Try alternate path
    try:
        sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
        from scan2doc_pro_v6 import BatchProcessor, QualityScorer, PageStatistics
    except ImportError:
        print("ERROR: Cannot import scan2doc_pro_v5.py")
        print("Make sure scan2doc_pro_v5.py is in the same directory as this script.")
        sys.exit(1)


# ═══════════════════════════════════════════════════════════════════════════════
# Color Palette
# ═══════════════════════════════════════════════════════════════════════════════

COLORS = {
    "bg_dark": "#1a1a2e",
    "bg_medium": "#16213e",
    "bg_light": "#0f3460",
    "accent": "#e94560",
    "accent2": "#533483",
    "text": "#eaeaea",
    "text_dim": "#8899aa",
    "success": "#00c853",
    "warning": "#ffab00",
    "error": "#ff1744",
    "sidebar": "#0f0f23",
    "card": "#1a1a3e",
    "blue": "#2979ff",
    "blue_dark": "#1565c0",
}


# ═══════════════════════════════════════════════════════════════════════════════
# File List Item Widget
# ═══════════════════════════════════════════════════════════════════════════════

class FileListItem(ctk.CTkFrame):
    """Individual file item in the file list."""

    def __init__(self, master, filepath, on_remove=None, **kwargs):
        super().__init__(master, fg_color=COLORS["card"], corner_radius=6, **kwargs)
        self.filepath = filepath
        self.on_remove = on_remove

        self.grid_columnconfigure(0, weight=1)

        # Filename label
        name = os.path.basename(filepath)
        if len(name) > 30:
            name = name[:27] + "..."
        self.label = ctk.CTkLabel(
            self, text=name, anchor="w",
            text_color=COLORS["text"], font=ctk.CTkFont(size=11))
        self.label.grid(row=0, column=0, padx=(8, 4), pady=4, sticky="ew")

        # Remove button
        self.remove_btn = ctk.CTkButton(
            self, text="✕", width=24, height=24,
            fg_color="transparent", hover_color=COLORS["error"],
            text_color=COLORS["text_dim"], font=ctk.CTkFont(size=12, weight="bold"),
            command=self._remove)
        self.remove_btn.grid(row=0, column=1, padx=(0, 6), pady=4)

    def _remove(self):
        if self.on_remove:
            self.on_remove(self.filepath)
        self.destroy()


# ═══════════════════════════════════════════════════════════════════════════════
# Sidebar Panel
# ═══════════════════════════════════════════════════════════════════════════════

class Sidebar(ctk.CTkFrame):
    """Left sidebar with file management and settings."""

    def __init__(self, master, app, **kwargs):
        super().__init__(master, fg_color=COLORS["sidebar"], corner_radius=0, width=280, **kwargs)
        self.app = app
        self.pack_propagate(False)
        self.grid_propagate(False)

        self.files = []

        self._build_ui()

    def _build_ui(self):
        # ── Logo / Title ──
        title_frame = ctk.CTkFrame(self, fg_color="transparent")
        title_frame.pack(fill="x", padx=15, pady=(15, 5))

        ctk.CTkLabel(
            title_frame, text="📄 Scan2Doc Pro",
            font=ctk.CTkFont(size=20, weight="bold"),
            text_color=COLORS["blue"]
        ).pack(anchor="w")

        ctk.CTkLabel(
            title_frame, text="Professional OCR → Word",
            font=ctk.CTkFont(size=11),
            text_color=COLORS["text_dim"]
        ).pack(anchor="w")

        # Separator
        ctk.CTkFrame(self, fg_color=COLORS["bg_light"], height=1).pack(fill="x", padx=15, pady=10)

        # ── File Selection ──
        ctk.CTkLabel(self, text="📁 Input Files", font=ctk.CTkFont(size=13, weight="bold"),
                     text_color=COLORS["text"]).pack(anchor="w", padx=15, pady=(5, 5))

        btn_frame = ctk.CTkFrame(self, fg_color="transparent")
        btn_frame.pack(fill="x", padx=15)

        ctk.CTkButton(
            btn_frame, text="📂 Select Folder", height=32,
            fg_color=COLORS["blue_dark"], hover_color=COLORS["blue"],
            font=ctk.CTkFont(size=12), command=self._select_folder
        ).pack(side="left", expand=True, fill="x", padx=(0, 4))

        ctk.CTkButton(
            btn_frame, text="📎 Add Files", height=32,
            fg_color=COLORS["accent2"], hover_color=COLORS["accent"],
            font=ctk.CTkFont(size=12), command=self._select_files
        ).pack(side="left", expand=True, fill="x", padx=(4, 0))

        # ── File List ──
        self.file_list_frame = ctk.CTkScrollableFrame(
            self, fg_color=COLORS["bg_medium"], corner_radius=8,
            height=120)
        self.file_list_frame.pack(fill="x", padx=15, pady=(8, 5))

        self.file_count_label = ctk.CTkLabel(
            self, text="No files selected", font=ctk.CTkFont(size=10),
            text_color=COLORS["text_dim"])
        self.file_count_label.pack(anchor="w", padx=15)

        # Clear all button
        ctk.CTkButton(
            self, text="🗑 Clear All", height=24, width=80,
            fg_color="transparent", border_width=1, border_color=COLORS["error"],
            text_color=COLORS["error"], hover_color="#330000",
            font=ctk.CTkFont(size=10), command=self._clear_files
        ).pack(anchor="e", padx=15, pady=(2, 8))

        # Separator
        ctk.CTkFrame(self, fg_color=COLORS["bg_light"], height=1).pack(fill="x", padx=15, pady=5)

        # ── Settings ──
        ctk.CTkLabel(self, text="⚙️ Settings", font=ctk.CTkFont(size=13, weight="bold"),
                     text_color=COLORS["text"]).pack(anchor="w", padx=15, pady=(5, 8))

        # DPI
        self._setting_label("DPI:")
        self.dpi_var = ctk.StringVar(value="300")
        ctk.CTkOptionMenu(
            self, variable=self.dpi_var,
            values=["150", "200", "300", "400", "500", "600"],
            fg_color=COLORS["bg_medium"], button_color=COLORS["blue_dark"],
            width=120, height=28, font=ctk.CTkFont(size=11)
        ).pack(anchor="w", padx=15, pady=(0, 6))

        # Language
        self._setting_label("Language:")
        self.lang_var = ctk.StringVar(value="Persian + English")
        ctk.CTkOptionMenu(
            self, variable=self.lang_var,
            values=["Persian + English", "Persian", "English", "Arabic", "All Languages"],
            fg_color=COLORS["bg_medium"], button_color=COLORS["blue_dark"],
            width=160, height=28, font=ctk.CTkFont(size=11)
        ).pack(anchor="w", padx=15, pady=(0, 6))

        # Title
        self._setting_label("Document Title:")
        self.title_entry = ctk.CTkEntry(
            self, placeholder_text="Optional title...",
            fg_color=COLORS["bg_medium"], border_color=COLORS["blue_dark"],
            height=28, font=ctk.CTkFont(size=11))
        self.title_entry.pack(fill="x", padx=15, pady=(0, 6))

        # Checkboxes
        self.tables_var = ctk.BooleanVar(value=True)
        self.lines_var = ctk.BooleanVar(value=True)
        self.images_var = ctk.BooleanVar(value=False)

        checks_frame = ctk.CTkFrame(self, fg_color="transparent")
        checks_frame.pack(fill="x", padx=15, pady=(0, 4))

        ctk.CTkCheckBox(
            checks_frame, text="Tables", variable=self.tables_var,
            text_color=COLORS["text"], font=ctk.CTkFont(size=11),
            fg_color=COLORS["blue_dark"], hover_color=COLORS["blue"],
            checkmark_color=COLORS["text"]
        ).pack(side="left", padx=(0, 8))

        ctk.CTkCheckBox(
            checks_frame, text="Lines", variable=self.lines_var,
            text_color=COLORS["text"], font=ctk.CTkFont(size=11),
            fg_color=COLORS["blue_dark"], hover_color=COLORS["blue"],
            checkmark_color=COLORS["text"]
        ).pack(side="left", padx=(0, 8))

        ctk.CTkCheckBox(
            checks_frame, text="Images", variable=self.images_var,
            text_color=COLORS["text"], font=ctk.CTkFont(size=11),
            fg_color=COLORS["blue_dark"], hover_color=COLORS["blue"],
            checkmark_color=COLORS["text"]
        ).pack(side="left")

        # Separator
        ctk.CTkFrame(self, fg_color=COLORS["bg_light"], height=1).pack(fill="x", padx=15, pady=8)

        # ── Output ──
        ctk.CTkLabel(self, text="💾 Output", font=ctk.CTkFont(size=13, weight="bold"),
                     text_color=COLORS["text"]).pack(anchor="w", padx=15, pady=(0, 5))

        self.output_var = ctk.StringVar(value="output.docx")
        self.output_entry = ctk.CTkEntry(
            self, textvariable=self.output_var,
            fg_color=COLORS["bg_medium"], border_color=COLORS["blue_dark"],
            height=28, font=ctk.CTkFont(size=11))
        self.output_entry.pack(fill="x", padx=15, pady=(0, 4))

        ctk.CTkButton(
            self, text="📂 Browse...", height=26,
            fg_color=COLORS["bg_medium"], border_width=1,
            border_color=COLORS["blue_dark"], hover_color=COLORS["blue_dark"],
            font=ctk.CTkFont(size=11), command=self._select_output
        ).pack(fill="x", padx=15, pady=(0, 10))

        # ── Start Button ──
        self.start_btn = ctk.CTkButton(
            self, text="▶ START PROCESSING", height=44,
            fg_color=COLORS["blue"], hover_color=COLORS["blue_dark"],
            font=ctk.CTkFont(size=15, weight="bold"),
            corner_radius=10, command=self._start_processing)
        self.start_btn.pack(fill="x", padx=15, pady=(0, 15))

    def _setting_label(self, text):
        ctk.CTkLabel(self, text=text, font=ctk.CTkFont(size=11),
                     text_color=COLORS["text_dim"]).pack(anchor="w", padx=15, pady=(0, 2))

    def _select_folder(self):
        try:
            from tkinter import filedialog
            folder = filedialog.askdirectory(title="Select Image Folder")
            if folder:
                exts = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif', '.webp', '.gif'}
                files = []
                for f in sorted(Path(folder).iterdir()):
                    if f.suffix.lower() in exts:
                        files.append(str(f))
                if files:
                    self.files.extend(files)
                    self._refresh_file_list()
                    self.app.log(f"Added {len(files)} files from folder")
                else:
                    self.app.log("No image files found in selected folder", "warning")
        except Exception as e:
            self.app.log(f"Error selecting folder: {e}", "error")

    def _select_files(self):
        try:
            from tkinter import filedialog
            files = filedialog.askopenfilenames(
                title="Select Image Files",
                filetypes=[
                    ("Image files", "*.jpg *.jpeg *.png *.bmp *.tiff *.tif *.webp *.gif"),
                    ("All files", "*.*")
                ])
            if files:
                self.files.extend(list(files))
                self._refresh_file_list()
                self.app.log(f"Added {len(files)} file(s)")
        except Exception as e:
            self.app.log(f"Error selecting files: {e}", "error")

    def _select_output(self):
        try:
            from tkinter import filedialog
            path = filedialog.asksaveasfilename(
                title="Save Output As",
                defaultextension=".docx",
                filetypes=[("Word Document", "*.docx")])
            if path:
                self.output_var.set(path)
        except Exception as e:
            self.app.log(f"Error: {e}", "error")

    def _clear_files(self):
        self.files.clear()
        self._refresh_file_list()
        self.app.log("File list cleared")

    def _remove_file(self, filepath):
        if filepath in self.files:
            self.files.remove(filepath)
        self._refresh_file_list()

    def _refresh_file_list(self):
        # Clear existing items
        for widget in self.file_list_frame.winfo_children():
            widget.destroy()

        for fp in self.files:
            item = FileListItem(self.file_list_frame, fp, on_remove=self._remove_file)
            item.pack(fill="x", padx=2, pady=2)

        count = len(self.files)
        self.file_count_label.configure(
            text=f"{count} file(s) selected" if count > 0 else "No files selected")

        # Auto-set output filename
        if count == 1:
            stem = Path(self.files[0]).stem
            self.output_var.set(f"{stem}.docx")
        elif count > 1:
            parent = os.path.basename(os.path.dirname(self.files[0]))
            if parent:
                self.output_var.set(f"{parent}.docx")

    def get_settings(self):
        """Return current settings as a dict."""
        lang_map = {
            "Persian + English": "fas+eng",
            "Persian": "fas",
            "English": "eng",
            "Arabic": "ara",
            "All Languages": "fas+ara+eng"
        }
        return {
            "files": self.files[:],
            "dpi": int(self.dpi_var.get()),
            "lang": lang_map.get(self.lang_var.get(), "fas+eng"),
            "title": self.title_entry.get().strip() or None,
            "output": self.output_var.get().strip() or "output.docx",
            "tables": self.tables_var.get(),
            "lines": self.lines_var.get(),
            "images": self.images_var.get(),
        }

    def set_processing_state(self, processing):
        """Enable/disable UI during processing."""
        state = "disabled" if processing else "normal"
        self.start_btn.configure(
            state=state,
            text="⏳ Processing..." if processing else "▶ START PROCESSING")


# ═══════════════════════════════════════════════════════════════════════════════
# Main Content Area
# ═══════════════════════════════════════════════════════════════════════════════

class MainArea(ctk.CTkFrame):
    """Right main area with preview, log, and progress."""

    def __init__(self, master, app, **kwargs):
        super().__init__(master, fg_color=COLORS["bg_dark"], corner_radius=0, **kwargs)
        self.app = app

        self.grid_rowconfigure(1, weight=1)
        self.grid_columnconfigure(0, weight=1)

        self._build_ui()

    def _build_ui(self):
        # ── Preview Panel ──
        preview_label = ctk.CTkLabel(
            self, text="🖼️ Preview", font=ctk.CTkFont(size=13, weight="bold"),
            text_color=COLORS["text"])
        preview_label.grid(row=0, column=0, padx=15, pady=(10, 5), sticky="w")

        self.preview_frame = ctk.CTkFrame(
            self, fg_color=COLORS["bg_medium"], corner_radius=10,
            height=250)
        self.preview_frame.grid(row=0, column=0, padx=15, pady=(30, 5), sticky="nsew")
        self.preview_frame.grid_propagate(False)

        self.preview_label = ctk.CTkLabel(
            self.preview_frame, text="Select files to preview",
            text_color=COLORS["text_dim"], font=ctk.CTkFont(size=12))
        self.preview_label.pack(expand=True)

        # ── Processing Log ──
        log_label = ctk.CTkLabel(
            self, text="📋 Processing Log", font=ctk.CTkFont(size=13, weight="bold"),
            text_color=COLORS["text"])
        log_label.grid(row=1, column=0, padx=15, pady=(10, 5), sticky="nw")

        self.log_text = ctk.CTkTextbox(
            self, fg_color=COLORS["bg_medium"], text_color=COLORS["text"],
            font=ctk.CTkFont(family="Consolas", size=11),
            corner_radius=10, wrap="word")
        self.log_text.grid(row=1, column=0, padx=15, pady=(30, 5), sticky="nsew")
        self.log_text.configure(state="disabled")

        # ── Footer: Progress + Status ──
        footer_frame = ctk.CTkFrame(self, fg_color=COLORS["bg_medium"], corner_radius=10)
        footer_frame.grid(row=2, column=0, padx=15, pady=(5, 10), sticky="ew")
        footer_frame.grid_columnconfigure(0, weight=1)

        # Progress bar
        self.progress_bar = ctk.CTkProgressBar(
            footer_frame, progress_color=COLORS["blue"],
            fg_color=COLORS["bg_dark"], height=12)
        self.progress_bar.grid(row=0, column=0, padx=10, pady=(8, 4), sticky="ew")
        self.progress_bar.set(0)

        # Status labels
        status_frame = ctk.CTkFrame(footer_frame, fg_color="transparent")
        status_frame.grid(row=1, column=0, padx=10, pady=(0, 8), sticky="ew")
        status_frame.grid_columnconfigure(0, weight=1)
        status_frame.grid_columnconfigure(1, weight=1)
        status_frame.grid_columnconfigure(2, weight=1)

        self.progress_label = ctk.CTkLabel(
            status_frame, text="0%", font=ctk.CTkFont(size=12, weight="bold"),
            text_color=COLORS["blue"])
        self.progress_label.grid(row=0, column=0, sticky="w")

        self.page_label = ctk.CTkLabel(
            status_frame, text="0 / 0 pages", font=ctk.CTkFont(size=11),
            text_color=COLORS["text_dim"])
        self.page_label.grid(row=0, column=1, sticky="w")

        self.quality_label = ctk.CTkLabel(
            status_frame, text="Quality: --", font=ctk.CTkFont(size=11, weight="bold"),
            text_color=COLORS["text_dim"])
        self.quality_label.grid(row=0, column=2, sticky="e")

    def log(self, message, level="info"):
        """Thread-safe log message."""
        prefix = {"info": "ℹ️", "warning": "⚠️", "error": "❌", "success": "✅"}.get(level, "ℹ️")
        color_map = {"info": COLORS["text"], "warning": COLORS["warning"],
                     "error": COLORS["error"], "success": COLORS["success"]}

        timestamp = time.strftime("%H:%M:%S")
        line = f"[{timestamp}] {prefix} {message}\n"

        def _append():
            self.log_text.configure(state="normal")
            self.log_text.insert("end", line)
            self.log_text.see("end")
            self.log_text.configure(state="disabled")

        try:
            self.after(0, _append)
        except Exception:
            pass

    def update_preview(self, filepath):
        """Update the preview panel with an image."""
        if not PIL_AVAILABLE:
            return

        def _update():
            try:
                for widget in self.preview_frame.winfo_children():
                    widget.destroy()

                img = Image.open(filepath)
                # Calculate fit size
                frame_w = self.preview_frame.winfo_width() - 20
                frame_h = self.preview_frame.winfo_height() - 20
                if frame_w < 10: frame_w = 400
                if frame_h < 10: frame_h = 200

                img.thumbnail((frame_w, frame_h), Image.Resampling.LANCZOS)
                photo = ImageTk.PhotoImage(img)

                label = ctk.CTkLabel(self.preview_frame, image=photo, text="")
                label.image = photo  # Keep reference
                label.pack(expand=True)
            except Exception as e:
                self.log(f"Preview error: {e}", "warning")

        try:
            self.after(0, _update)
        except Exception:
            pass

    def update_progress(self, current, total, message=""):
        """Update progress bar and labels."""
        def _update():
            pct = (current / total * 100) if total > 0 else 0
            self.progress_bar.set(current / total if total > 0 else 0)
            self.progress_label.configure(text=f"{pct:.0f}%")
            self.page_label.configure(text=f"{current} / {total} pages")

        try:
            self.after(0, _update)
        except Exception:
            pass

    def update_quality(self, score):
        """Update quality score display."""
        def _update():
            if score >= 9:
                color = COLORS["success"]
            elif score >= 7:
                color = COLORS["blue"]
            elif score >= 5:
                color = COLORS["warning"]
            else:
                color = COLORS["error"]
            self.quality_label.configure(text=f"Quality: {score}/10", text_color=color)

        try:
            self.after(0, _update)
        except Exception:
            pass

    def reset(self):
        """Reset progress display."""
        self.progress_bar.set(0)
        self.progress_label.configure(text="0%", text_color=COLORS["blue"])
        self.page_label.configure(text="0 / 0 pages")
        self.quality_label.configure(text="Quality: --", text_color=COLORS["text_dim"])


# ═══════════════════════════════════════════════════════════════════════════════
# Main Application Window
# ═══════════════════════════════════════════════════════════════════════════════

class Scan2DocApp(ctk.CTk):
    """Main application window."""

    def __init__(self):
        super().__init__()

        self.title("Scan2Doc Pro — Professional OCR to Word")
        self.geometry("1280x800")
        self.minsize(900, 600)
        self.configure(fg_color=COLORS["bg_dark"])

        self.processing = False
        self._build_layout()

    def _build_layout(self):
        # Grid layout: sidebar (fixed) + main area (expandable)
        self.grid_columnconfigure(1, weight=1)
        self.grid_rowconfigure(0, weight=1)

        # Sidebar
        self.sidebar = Sidebar(self, self)
        self.sidebar.grid(row=0, column=0, sticky="ns")

        # Main area
        self.main_area = MainArea(self, self)
        self.main_area.grid(row=0, column=1, sticky="nsew")

    def log(self, message, level="info"):
        """Log a message to the processing log."""
        self.main_area.log(message, level)

    def _start_processing(self):
        """Start OCR processing in a background thread."""
        settings = self.sidebar.get_settings()

        if not settings["files"]:
            self.log("No files selected! Please add images first.", "error")
            return

        if self.processing:
            return

        self.processing = True
        self.sidebar.set_processing_state(True)
        self.main_area.reset()

        # Show first image preview
        if settings["files"]:
            self.main_area.update_preview(settings["files"][0])

        thread = threading.Thread(target=self._process_worker, args=(settings,), daemon=True)
        thread.start()

    def _process_worker(self, settings):
        """Background processing worker."""
        try:
            self.log(f"Starting processing: {len(settings['files'])} file(s)", "info")
            self.log(f"Settings: DPI={settings['dpi']}, Lang={settings['lang']}", "info")

            t0 = time.time()

            def progress_callback(current, total, msg):
                self.main_area.update_progress(current, total, msg)
                self.log(f"Page {current}/{total}: {msg}", "info")

            processor = BatchProcessor(
                lang=settings["lang"],
                dpi=settings["dpi"],
                title=settings["title"],
                progress_callback=progress_callback
            )

            # Suppress print output, capture via log
            import io
            old_stdout = sys.stdout
            sys.stdout = io.StringIO()

            ok = processor.process_batch(
                settings["files"],
                settings["output"],
                title=settings["title"]
            )

            sys.stdout = old_stdout

            # Calculate quality score
            if processor.stats:
                ac = sum(s.overall_confidence for s in processor.stats) / len(processor.stats)
                tb = sum(s.text_blocks for s in processor.stats)
                tt = sum(s.tables for s in processor.stats)
                tl = sum(s.lines for s in processor.stats)
                tw = sum(s.total_words for s in processor.stats)
                lc = sum(s.low_confidence_words for s in processor.stats)

                agg_stats = {
                    'tables_found': tt, 'confidence': ac, 'blocks': tb,
                    'low_conf_words': lc, 'lines_found': tl,
                    'columns': max((s.columns for s in processor.stats), default=1),
                    'word_count': tw,
                }
                quality = QualityScorer.calculate(agg_stats)
                self.main_area.update_quality(quality)
                self.log(f"Quality Score: {quality}/10", "success")
                self.log(f"Blocks: {tb}, Tables: {tt}, Words: {tw}, Confidence: {ac:.1f}%", "info")

            elapsed = time.time() - t0
            if ok:
                self.log(f"Done! Output saved to: {settings['output']}", "success")
                self.log(f"Total time: {elapsed:.1f}s", "info")
                self.main_area.update_progress(len(settings["files"]), len(settings["files"]))
            else:
                self.log("Processing failed. Check the log for errors.", "error")

        except Exception as e:
            self.log(f"Processing error: {e}", "error")
            self.log(traceback.format_exc(), "error")
            sys.stdout = sys.__stdout__
        finally:
            self.processing = False
            try:
                self.after(0, lambda: self.sidebar.set_processing_state(False))
            except Exception:
                pass


# ═══════════════════════════════════════════════════════════════════════════════
# Entry Point
# ═══════════════════════════════════════════════════════════════════════════════

def main():
    """Launch the application."""
    app = Scan2DocApp()
    app.mainloop()


if __name__ == "__main__":
    main()