cv.py

import cv2
import numpy as np
from sklearn.neighbors import NearestNeighbors
import os

def detect_back_patches(image_path, threshold_distance, save_path):
    # Read the image
    image = cv2.imread(image_path)
    
    # Convert image to grayscale if necessary
    gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    
    # Apply nearest neighbor algorithm
    # Assuming patches are represented as feature vectors
    patches = extract_patches(gray_image)  # Extract patches from the image
    nn = NearestNeighbors(n_neighbors=2, algorithm='auto')
    nn.fit(patches)
    
    # Find nearest neighbors
    distances, indices = nn.kneighbors(patches)
    
    # Filter patches based on threshold distance
    filtered_patches_indices = np.where(distances[:,1] > threshold_distance)[0]
    
    # Get the ROIs corresponding to the filtered patches
    ROIs = [patches[idx] for idx in filtered_patches_indices]
    
    # Save the black patches as images
    if not os.path.exists(save_path):
        os.makedirs(save_path)
    
    for i, roi in enumerate(ROIs):
        patch_image = roi.reshape(gray_image.shape[0] // 32, gray_image.shape[1] // 32)
        cv2.imwrite(os.path.join(save_path, f"black_patch_{i}.jpg"), patch_image)
    
    return ROIs

def extract_patches(image, patch_size=(32, 32)):
    # Extract patches from the image
    patches = []
    height, width = image.shape[:2]
    patch_height, patch_width = patch_size
    
    for y in range(0, height - patch_height + 1, patch_height):
        for x in range(0, width - patch_width + 1, patch_width):
            patch = image[y:y+patch_height, x:x+patch_width]
            patches.append(patch.flatten())  # Flatten patch to create feature vector
    
    return np.array(patches)

# Example usage
image_path = 'document_image.jpg'
threshold_distance = 100  # Adjust this threshold based on your requirements
save_path = 'black_patches'
back_patches_ROIs = detect_back_patches(image_path, threshold_distance, save_path)