We load the collage image, locate and crop the two panels, align their sizes, compute absolute differences, suppress noise with Gaussian blur, and detect contours over the changed regions.
### Import OpenCV for image processing and computer vision operations. import cv2 ### Import NumPy for numerical array operations and mask handling. import numpy as np ### Read the collage that contains two side-by-side images into a NumPy array. img = cv2 . imread ( " Open-CV/Find-10-Differences/two-images.jpg " ) ### Convert the image to grayscale to simplify thresholding and edge detection. gray = cv2 . cvtColor ( img , cv2 . COLOR_RGB2GRAY ) ### Apply a binary inverse threshold to separate the two image panels from the background. _ , thresh = cv2 . threshold ( gray , 230 , 255 , cv2 . THRESH_BINARY_INV ) ### Run Canny edge detection to get crisp borders around the two panels. canny = cv2 . Canny ( thresh , 254 , 255 ) ### Find external contours to capture the two largest rectangular regions that contain the images. contours , hierarchy = cv2 . findContours ( canny . copy () , cv2 . RETR_EXTERNAL , cv2 . CHAIN_APPROX_SIMPLE ) ### Optionally draw contours for debugging to verify panel detection accuracy. #cv2.drawContours(img, contours, -1 , (0,255,0) ,2 ) ### Sort contours by area so we can retrieve the two biggest panels reliably. sortedCon = sorted ( contours , key = cv2 . contourArea ) ### Compute a bounding rectangle around the largest panel. x , y , w , h = cv2 . boundingRect ( sortedCon [ 0 ]) ### Draw a rectangle on the collage to visualize the first panel boundaries. cv2 . rectangle ( img , ( x , y ), ( x + w , y + h ), ( 255 , 0 , 0 ), 3 ) ### Crop the first region of interest (ROI) that represents the first image. roi = img . copy ()[ y : y + h , x : x + w ] ### Compute a bounding rectangle around the second panel. x1 , y1 , w1 , h1 = cv2 . boundingRect ( sortedCon [ 1 ]) ### Draw a rectangle to visualize the second panel boundaries. cv2 . rectangle ( img , ( x1 , y1 ), ( x1 + w1 , y1 + h1 ), ( 255 , 0 , 0 ), 3 ) ### Crop the second region of interest (ROI) that represents the second image. roi2 = img . copy ()[ y1 : y1 + h1 , x1 : x1 + w1 ] ### Print the shapes so we can confirm both ROIs and plan alignment if needed. print ( roi . shape ) ### Print the second ROI shape for comparison against the first ROI. print ( roi2 . shape ) ### Retrieve the height of the second ROI to trim one row if necessary for shape alignment. roi2H = roi2 . shape [ 0 ] ### Print the current height for transparency during debugging. print ( roi2H ) ### Align heights by removing a bottom row if the sizes are off by one pixel. roi2 = roi2 [ 0 : roi2H - 1 , :] ### Confirm the new shape after alignment. print ( roi2 . shape ) ### Retrieve the width of the first ROI to trim one column if necessary for shape alignment. roi1W = roi . shape [ 1 ] ### Print the current width for transparency during debugging. print ( roi1W ) ### Align widths by removing the last column if the sizes are off by one pixel. roi = roi [: , 0 : roi1W - 1 ] ### Confirm the new shape after alignment. print ( roi . shape ) ### Compute pixel-wise absolute difference and invert to emphasize similar background while keeping differences bright. diff = 255 - cv2 . absdiff ( roi , roi2 ) ### Define the lower HSV threshold for masking potential difference regions. lower = np . array ([ 0 , 0 , 0 ]) ### Define the upper HSV threshold to include the full range of values. upper = np . array ([ 180 , 255 , 255 ]) ### Convert the difference image from BGR to HSV for robust value-based masking. imgHSV = cv2 . cvtColor ( diff , cv2 . COLOR_BGR2HSV ) ### Create a binary mask selecting all HSV values in the defined range. mask = cv2 . inRange ( imgHSV , lower , upper ) ### Keep only the masked regions from the difference image for cleaner downstream processing. result = cv2 . bitwise_and ( diff , diff , mask = mask ) ### Apply Gaussian blur to reduce salt-and-pepper noise before edge detection. blur = cv2 . GaussianBlur ( result , ( 11 , 11 ), 0 ) ### Run Canny a second time on the denoised image to outline difference boundaries sharply. canny = cv2 . Canny ( blur , 50 , 150 ) ### Dilate the edges to close small gaps and make contours easier to detect. edges = cv2 . dilate ( canny , None ) ### Extract contours corresponding to distinct difference regions between the two images. contours , hierarchy = cv2 . findContours ( edges , cv2 . RETR_EXTERNAL , cv2 . CHAIN_APPROX_SIMPLE ) ### Iterate over each contour to compute bounding boxes around detected differences. for cnt in contours : ### Get the top-left coordinate and size of the bounding rectangle for the current difference region. x , y , width , height = cv2 . boundingRect ( cnt ) ### Draw the bounding box on the result canvas to visualize detected differences. cv2 . rectangle ( result , ( x , y ), ( x + width , y + height ), ( 0 , 0 , 255 ), 2 ) ### Draw the same bounding box on the first ROI for side-by-side visualization. cv2 . rectangle ( roi , ( x , y ), ( x + width , y + height ), ( 0 , 0 , 255 ), 2 ) ### Draw the same bounding box on the second ROI for side-by-side visualization. cv2 . rectangle ( roi2 , ( x , y ), ( x + width , y + height ), ( 0 , 0 , 255 ), 2 ) ### Display the composite result with highlighted differences. cv2 . imshow ( " result " , result ) ### Display the first cropped image with bounding boxes. cv2 . imshow ( " roi " , roi ) ### Display the second cropped image with bounding boxes. cv2 . imshow ( " roi2 " , roi2 ) ### Display the Canny output used to detect edges of the difference regions. cv2 . imshow ( " canny " , canny ) ### Display the dilated edges that feed the contour extraction. cv2 . imshow ( " edges " , edges ) ### Wait for a key press to keep windows open until the user is ready to close them. cv2 . waitKey ( 0 ) ### Close all OpenCV windows to release resources gracefully. cv2 . destroyAllWindows () 
Skip to content 
