From 21c871d50117e3362432d8e133bce99315d0e2de Mon Sep 17 00:00:00 2001
From: Siddharth Kumar Sah <siddharth123sk@gmail.com>
Date: Sun, 12 Apr 2026 18:30:00 +0800
Subject: [PATCH] feat(ocr): add OpenCV preprocessing pipeline for OCR

---
 packages/ai/python/ocr_preprocess.py | 86 ++++++++++++++++++++++++++++
 1 file changed, 86 insertions(+)
 create mode 100644 packages/ai/python/ocr_preprocess.py

diff --git a/packages/ai/python/ocr_preprocess.py b/packages/ai/python/ocr_preprocess.py
new file mode 100644
index 00000000..7b67540f
--- /dev/null
+++ b/packages/ai/python/ocr_preprocess.py
@@ -0,0 +1,86 @@
+"""Image preprocessing pipeline for OCR accuracy improvement.
+
+Uses OpenCV for deskew, adaptive binarization, CLAHE contrast enhancement,
+and denoising. All operations work on the image file in-place (overwrite).
+"""
+import cv2
+import numpy as np
+import sys
+import json
+
+
+def emit_progress(percent, stage):
+    print(json.dumps({"progress": percent, "stage": stage}), file=sys.stderr, flush=True)
+
+
+def deskew(image):
+    """Detect and correct rotation/skew using Hough line transform."""
+    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if len(image.shape) == 3 else image
+    edges = cv2.Canny(gray, 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.degrees(np.arctan2(y2 - y1, x2 - x1))
+        if abs(angle) < 45:
+            angles.append(angle)
+
+    if not angles:
+        return image
+
+    median_angle = np.median(angles)
+    if abs(median_angle) < 0.5 or abs(median_angle) > 15:
+        return image
+
+    h, w = image.shape[:2]
+    center = (w // 2, h // 2)
+    matrix = cv2.getRotationMatrix2D(center, median_angle, 1.0)
+    rotated = cv2.warpAffine(image, matrix, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
+    return rotated
+
+
+def binarize(image):
+    """Adaptive thresholding for high-contrast black/white."""
+    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if len(image.shape) == 3 else image
+    binary = cv2.adaptiveThreshold(
+        gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2
+    )
+    return cv2.cvtColor(binary, cv2.COLOR_GRAY2BGR)
+
+
+def enhance_contrast(image):
+    """CLAHE contrast enhancement for uneven lighting."""
+    lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)
+    l, a, b = cv2.split(lab)
+    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+    l = clahe.apply(l)
+    enhanced = cv2.merge([l, a, b])
+    return cv2.cvtColor(enhanced, cv2.COLOR_LAB2BGR)
+
+
+def denoise(image):
+    """Bilateral filter to remove noise while preserving text edges."""
+    return cv2.bilateralFilter(image, 9, 75, 75)
+
+
+def preprocess(input_path, output_path):
+    """Run the full preprocessing pipeline and save result.
+
+    Steps: deskew -> enhance contrast -> denoise -> binarize
+    Order matters: binarize last because it strips color info needed by CLAHE.
+    """
+    image = cv2.imread(input_path)
+    if image is None:
+        raise ValueError(f"Cannot read image: {input_path}")
+
+    image = deskew(image)
+    image = enhance_contrast(image)
+    image = denoise(image)
+    image = binarize(image)
+
+    cv2.imwrite(output_path, image)
+    return output_path