main.py

from fastapi import FastAPI, UploadFile, File, Form
from doctr.io import DocumentFile
from doctr.models import ocr_predictor
from PIL import Image, ImageOps, ImageEnhance, ImageFilter
import tempfile
import io
import os
import numpy as np
import cv2

app = FastAPI()

print("Loading docTR OCR model...")
doctr_model = ocr_predictor(pretrained=True)
print("docTR ready.")

print("Loading TrOCR handwriting model...")
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
trocr_processor = TrOCRProcessor.from_pretrained("microsoft/trocr-large-handwritten")
trocr_model     = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-large-handwritten")
print("TrOCR ready.")


def fix_image_orientation(img: Image.Image) -> Image.Image:
    """Fix EXIF rotation AND force portrait if landscape"""
    try:
        img = ImageOps.exif_transpose(img)
    except Exception:
        pass
    # ✅ Force portrait — landscape photos rotated 90°
    w, h = img.size
    if w > h:
        img = img.rotate(90, expand=True)
    return img


def deskew(img: Image.Image) -> Image.Image:
    """Auto-correct small tilts using OpenCV"""
    try:
        cv_img = np.array(img)
        gray   = cv2.cvtColor(cv_img, cv2.COLOR_RGB2GRAY)
        coords = np.column_stack(np.where(gray < 200))
        if len(coords) > 0:
            angle = cv2.minAreaRect(coords)[-1]
            if angle < -45:
                angle = 90 + angle
            if abs(angle) > 0.5:
                (h, w) = cv_img.shape[:2]
                center = (w // 2, h // 2)
                M      = cv2.getRotationMatrix2D(center, angle, 1.0)
                cv_img = cv2.warpAffine(cv_img, M, (w, h),
                                        flags=cv2.INTER_CUBIC,
                                        borderMode=cv2.BORDER_REPLICATE)
        return Image.fromarray(cv_img)
    except Exception:
        return img


def enhance_image(img: Image.Image) -> Image.Image:
    img = deskew(img)
    img = ImageEnhance.Contrast(img).enhance(1.4)
    img = ImageEnhance.Sharpness(img).enhance(2.0)
    img = ImageEnhance.Brightness(img).enhance(1.1)
    return img


def enhance_for_handwriting(img: Image.Image) -> Image.Image:
    img = deskew(img)
    img = img.convert("L")
    img = ImageEnhance.Contrast(img).enhance(2.0)
    img = ImageEnhance.Sharpness(img).enhance(2.5)
    img = img.filter(ImageFilter.SHARPEN)
    img = img.convert("RGB")
    return img


def run_trocr(img: Image.Image) -> str:
    """Run Microsoft TrOCR for handwritten images — processes in chunks"""
    import torch

    width, height = img.size
    chunk_height  = 100
    results       = []

    for y in range(0, height, chunk_height):
        crop = img.crop((0, y, width, min(y + chunk_height, height)))
        if crop.size[1] < 20:
            continue
        pixel_values = trocr_processor(images=crop, return_tensors="pt").pixel_values
        with torch.no_grad():
            generated_ids = trocr_model.generate(pixel_values)
        text = trocr_processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
        if text:
            results.append(text)

    return "\n".join(results)


@app.get("/")
def root():
    return {"status": "OCR running"}


@app.post("/ocr")
async def ocr_images(
    file: UploadFile = File(...),
    mode: str = Form("print")
):
    try:
        contents  = await file.read()
        pil_image = Image.open(io.BytesIO(contents)).convert("RGB")
        pil_image = fix_image_orientation(pil_image)  # ✅ fixes landscape + EXIF

        # ── HANDWRITING MODE ──────────────────────────────────────
        if mode == "handwrite":
            pil_image = enhance_for_handwriting(pil_image)
            text      = run_trocr(pil_image)

        # ── PRINT MODE ────────────────────────────────────────────
        elif mode == "print":
            pil_image = enhance_image(pil_image)
            with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp:
                pil_image.save(tmp.name, quality=95)
                temp_path = tmp.name
            doc    = DocumentFile.from_images([temp_path])
            result = doctr_model(doc)
            text   = result.render().strip()
            os.remove(temp_path)

        # ── AUTO MODE ────────────────────────────────────────────
        else:
            pil_enhanced = enhance_image(pil_image.copy())
            with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp:
                pil_enhanced.save(tmp.name, quality=95)
                temp_path = tmp.name
            doc    = DocumentFile.from_images([temp_path])
            result = doctr_model(doc)
            text   = result.render().strip()
            os.remove(temp_path)

            if len(text) < 150:
                pil_hw = enhance_for_handwriting(pil_image.copy())
                text   = run_trocr(pil_hw)

        if not text or len(text) < 10:
            return {
                "success": False,
                "error": "No text found in image. Try a clearer photo with visible text."
            }

        return {"success": True, "text": text}

    except Exception as e:
        return {"success": False, "error": str(e)}