extract.py

import pymupdf  # PyMuPDF
import easyocr
import io
import os

# Initialise EasyOCR once (French + English) — outside the function for efficiency
# First run will download models (~400 MB); subsequent runs are fast.
reader = easyocr.Reader(['fr', 'en'], gpu=False)  # Set gpu=True if you have CUDA

def extract_pdf_text_with_easyocr(
    file_path: str,
    zoom: float = 3.0,           # 3.0 ≈ 300 DPI — good balance of quality/speed
    min_text_length: int = 50,   # Ignore pages with very little text (likely blank)
    save_to_file: str = None     # Optional: path to save extracted text
) -> str:
    """
    Extract text from a PDF using EasyOCR with OCR fallback for image-based pages.
    Returns clean concatenated text.
    """
    if not os.path.exists(file_path):
        raise FileNotFoundError(f"PDF file not found: {file_path}")

    try:
        doc = pymupdf.open(file_path)
    except Exception as e:
        raise RuntimeError(f"Failed to open PDF with PyMuPDF: {e}")

    full_text = ""
    total_pages = len(doc)

    for page_num in range(total_pages):
        page = doc[page_num]

        # Render page to high-resolution pixmap
        mat = pymupdf.Matrix(zoom, zoom)
        pix = page.get_pixmap(matrix=mat, colorspace=pymupdf.csRGB)

        # Convert to PNG bytes
        img_bytes = pix.tobytes("png")

        # Perform OCR
        try:
            result = reader.readtext(
                img_bytes,
                detail=0,           # Return only text strings
                paragraph=True,     # Group into paragraphs
                width_ths=0.7,      # Adjust for better line grouping
                height_ths=0.7
            )
            page_text = "\n".join([line.strip() for line in result if line.strip()])
        except Exception as ocr_error:
            page_text = f"[OCR Error on page {page_num + 1}: {ocr_error}]"

        # Only add page if meaningful text was extracted
        if len(page_text) > min_text_length or "OCR Error" in page_text:
            full_text += f"--- Page {page_num + 1} ---\n{page_text}\n\n"

    doc.close()

    # Optional: Save to file
    if save_to_file:
        try:
            with open(save_to_file, "w", encoding="utf-8") as f:
                f.write(full_text)
            print(f"Extracted text saved to: {save_to_file}")
        except Exception as e:
            print(f"Failed to save file: {e}")

    return full_text.strip()

# ———————— TEST USAGE ————————
if __name__ == "__main__":
    pdf_file = "Kbis.pdf"  # Update with your actual filename/path

    extracted_text = extract_pdf_text_with_easyocr(
        file_path=pdf_file,
        zoom=3.5,                    # Slightly higher for very small text
        save_to_file="kbis_extracted.txt"
    )

    # Print preview (first 1000 chars)
    print("\n=== EXTRACTED TEXT PREVIEW ===\n")
    print(extracted_text[:1000])
    print("\n... (truncated)" if len(extracted_text) > 1000 else "")