Spaces:
Running
Running
File size: 4,057 Bytes
e56eb98 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 | """
Preprocessing utilities for marksheet images.
Provides a simple pipeline:
- Load image
- Detect the largest page-like contour and crop so only the marksheet is visible
- Boost contrast using CLAHE on the L (or V) channel
- Reduce saturation slightly to stabilize OCR
Returns (processed_image, original_image, crop_coords)
where crop_coords = (x1, y1, x2, y2) in original image coordinates.
"""
from typing import Tuple, Optional
import cv2
import numpy as np
def _find_largest_contour_cropping_box(image_gray: np.ndarray) -> Optional[Tuple[int, int, int, int]]:
"""
Find a tight bounding box around the largest high-contrast region (the marksheet page).
This uses Canny + contour detection and returns a bounding rectangle.
"""
# Edges
blurred = cv2.GaussianBlur(image_gray, (5, 5), 0)
edges = cv2.Canny(blurred, 50, 150)
# Dilate to connect edges
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
edges = cv2.dilate(edges, kernel, iterations=1)
contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if not contours:
return None
# Choose the largest contour by area
largest_cnt = max(contours, key=cv2.contourArea)
x, y, w, h = cv2.boundingRect(largest_cnt)
# Expand slightly to avoid tight cuts
pad_x = int(0.01 * image_gray.shape[1])
pad_y = int(0.01 * image_gray.shape[0])
x1 = max(0, x - pad_x)
y1 = max(0, y - pad_y)
x2 = min(image_gray.shape[1], x + w + pad_x)
y2 = min(image_gray.shape[0], y + h + pad_y)
return (x1, y1, x2, y2)
def _apply_contrast_and_saturation(img_bgr: np.ndarray) -> np.ndarray:
"""
Boost contrast with CLAHE on the L channel (via LAB) and reduce saturation ~15% via HSV.
"""
# Contrast (CLAHE on LAB L channel)
lab = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2LAB)
l, a, b = cv2.split(lab)
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
l_eq = clahe.apply(l)
lab_eq = cv2.merge((l_eq, a, b))
bgr_eq = cv2.cvtColor(lab_eq, cv2.COLOR_LAB2BGR)
# Reduce saturation in HSV
hsv = cv2.cvtColor(bgr_eq, cv2.COLOR_BGR2HSV)
h, s, v = cv2.split(hsv)
s = (s.astype(np.float32) * 0.85).clip(0, 255).astype(np.uint8)
hsv_mod = cv2.merge((h, s, v))
bgr_out = cv2.cvtColor(hsv_mod, cv2.COLOR_HSV2BGR)
return bgr_out
def preprocess_marksheet(image_path: str,
output_path: Optional[str] = None,
save_intermediate: bool = False):
"""
Load an image, crop to the marksheet region, enhance contrast and reduce saturation.
Args:
image_path: Path to the input image
output_path: Optional path to save the processed image (unused when None)
save_intermediate: Whether to save intermediate images (not used; kept for compatibility)
Returns:
processed_image (np.ndarray), original_image (np.ndarray), crop_coords (x1,y1,x2,y2)
"""
original = cv2.imread(image_path)
if original is None:
raise FileNotFoundError(f"Failed to read image: {image_path}")
# Ensure portrait orientation similar to downstream expectations
h, w = original.shape[:2]
img = original.copy()
if w > h:
img = cv2.rotate(img, cv2.ROTATE_90_COUNTERCLOCKWISE)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Find crop box
crop_box = _find_largest_contour_cropping_box(gray)
if crop_box is None:
# Fall back to the whole image
x1, y1, x2, y2 = 0, 0, img.shape[1], img.shape[0]
else:
x1, y1, x2, y2 = crop_box
cropped = img[y1:y2, x1:x2]
if cropped.size == 0:
cropped = img
x1, y1, x2, y2 = 0, 0, img.shape[1], img.shape[0]
# Enhance
enhanced = _apply_contrast_and_saturation(cropped)
# Save if path provided
if output_path:
try:
cv2.imwrite(output_path, enhanced)
except Exception:
pass
crop_coords = (int(x1), int(y1), int(x2), int(y2))
return enhanced, original, crop_coords
|