roa2web-service-auto/data-entry-app/backend/app/services/image_preprocessor.py

"""Image preprocessing for optimal OCR results."""

from pathlib import Path
from typing import List

import numpy as np
import cv2

try:
    import pdf2image
    PDF_AVAILABLE = True
except ImportError:
    PDF_AVAILABLE = False


class ImagePreprocessor:
    """Preprocess receipt images for OCR."""

    def load_image(self, path: Path) -> np.ndarray:
        """Load image from file."""
        image = cv2.imread(str(path))
        if image is None:
            raise ValueError(f"Could not load image: {path}")
        return image

    def pdf_to_images(self, path: Path, dpi: int = 300) -> List[np.ndarray]:
        """
        Convert PDF to images.

        Args:
            path: Path to PDF file
            dpi: Resolution (300 = fast & good quality, 400 = better but slower)
        """
        if not PDF_AVAILABLE:
            raise RuntimeError("pdf2image not available. Install with: pip install pdf2image")
        images = pdf2image.convert_from_path(str(path), dpi=dpi)
        return [np.array(img) for img in images]

    def preprocess(self, image: np.ndarray, high_quality: bool = True) -> np.ndarray:
        """
        Apply LIGHT preprocessing - better for clear PDFs.
        Heavy binarization can destroy text on clear images.
        """
        return self.preprocess_light(image)

    def preprocess_light(self, image: np.ndarray) -> np.ndarray:
        """
        Light preprocessing for CLEAR images (PDFs, good scans).
        Preserves original quality, only enhances contrast.
        """
        # 1. Grayscale
        if len(image.shape) == 3:
            gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        else:
            gray = image.copy()

        # 2. Resize if too small
        height, width = gray.shape
        if width < 1500:
            scale = 1500 / width
            gray = cv2.resize(gray, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)

        # 3. Deskew
        gray = self._deskew(gray)

        # 4. Light contrast enhancement only
        clahe = cv2.createCLAHE(clipLimit=1.5, tileGridSize=(8, 8))
        enhanced = clahe.apply(gray)

        # NO binarization, NO morphological ops - preserve original quality
        return enhanced

    def preprocess_heavy(self, image: np.ndarray) -> np.ndarray:
        """
        Heavy preprocessing for FADED thermal receipts.
        Aggressive binarization to recover faded text.
        """
        # 1. Grayscale
        if len(image.shape) == 3:
            gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        else:
            gray = image.copy()

        # 2. Resize if too small (larger = better OCR)
        height, width = gray.shape
        if width < 1500:
            scale = 1500 / width
            gray = cv2.resize(gray, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)

        # 3. Deskew
        gray = self._deskew(gray)

        # 4. Contrast enhancement with CLAHE
        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
        enhanced = clahe.apply(gray)

        # 5. Denoise
        denoised = cv2.fastNlMeansDenoising(enhanced, h=8, templateWindowSize=7, searchWindowSize=21)

        # 6. Sharpening
        gaussian = cv2.GaussianBlur(denoised, (0, 0), 2.0)
        sharpened = cv2.addWeighted(denoised, 1.5, gaussian, -0.5, 0)

        # 7. Adaptive thresholding (binarization)
        binary = cv2.adaptiveThreshold(
            sharpened, 255,
            cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
            cv2.THRESH_BINARY,
            blockSize=11, C=5
        )

        # 8. Morphological operations
        kernel_close = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
        result = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel_close)

        return result

    def get_all_variants(self, image: np.ndarray) -> List[np.ndarray]:
        """
        Generate 2 preprocessing variants for OCR (fast mode).
        Returns: [light_processed, heavy_processed]
        """
        return [
            self.preprocess_light(image),
            self.preprocess_heavy(image),
        ]

    def _deskew(self, image: np.ndarray) -> np.ndarray:
        """Correct image rotation/skew using Hough lines."""
        edges = cv2.Canny(image, 50, 150, apertureSize=3)
        lines = cv2.HoughLinesP(
            edges, 1, np.pi / 180,
            threshold=100, minLineLength=100, maxLineGap=10
        )

        if lines is None:
            return image

        angles = []
        for line in lines:
            x1, y1, x2, y2 = line[0]
            angle = np.arctan2(y2 - y1, x2 - x1) * 180 / np.pi
            if abs(angle) < 45:
                angles.append(angle)

        if not angles:
            return image

        median_angle = np.median(angles)
        if abs(median_angle) < 0.5:
            return image

        h, w = image.shape[:2]
        center = (w // 2, h // 2)
        M = cv2.getRotationMatrix2D(center, median_angle, 1.0)
        return cv2.warpAffine(
            image, M, (w, h),
            flags=cv2.INTER_CUBIC,
            borderMode=cv2.BORDER_REPLICATE
        )