Compare two images and find all pixel coordinates that differ

user_16368306 0
Viewed 965

I have designed a program that compares two images and gives you the coordinates of the pixels that are different in both images and plots the using pygame. I do not mind having to use another library or to remake my whole code but it should ideally take less that 0.6s to process and it should not reduce file size, all I need it to do is to return the coordinates relative to the image

My code:

import cv2
import pygame
from pygame.locals import *
import time

lib = 'Map1.png'
lib2 = 'Map2.png'
lib3 = ()
coordenatesx = ()
coordenatesy = ()

Read = list(cv2.imread(lib).astype("int"))
Read2 = list(cv2.imread(lib2).astype("int"))

counter = 0

pygame.init()

flags = DOUBLEBUF
screen = pygame.display.set_mode((500,500), flags, 4)

start = time.process_time()#To tell me how long it takes
for y in range(len(Read)):#y coords
    for x in range(len(Read[y])):#x coords
        all = list(Read[y][x])[0]
        all2 = list(Read2[y][x])[0]
        difference = (all)-(all2)
        if difference > 10 or difference < -10: #To see if the pixel's difference is in the boundary if not it is different and it gets plotted
            counter+=1
            pygame.draw.rect(screen, (255, 0, 0), pygame.Rect(x, y, 1, 1))

pygame.display.update()
print(time. process_time() - start)

if counter >= (y * x) * 0.75:
    print('They are similar images')
    print('They are different by only :', str((counter / (y * x)) * 100), '%')
else:
    print('They are different')
    print('They are different by:', str((counter / (y * x)) * 100), '%')

pygame.display.update()
image1 image2
image_1 image_2
computer-vision opencv image-processing image python
2 Answers

You do not need to use for loop to do the same.

Numpy makes things simple:

  • it's easy to understand
  • speed up operations

Reading both your images in grayscale:

img1 = cv2.imread(r'C:\Users\524316\Desktop\Stack\m1.png', 0)
1mg2 = cv2.imread(r'C:\Users\524316\Desktop\Stack\m2.png', 0)

Subtracting them. cv2.subtract() takes care of normalization such that it doesn't return negative values. Coordinates having no change remain black (pixel intensity = 0)

sub = cv2.subtract(img1, img2)

Using numpy find the coordinates where changes are more than 0

coords = np.argwhere(sub > 0)

# return first 10 elements of the array coords
coords[:10]

array([[ 0, 23],
       [ 0, 24],
       [ 0, 25],
       [ 0, 26],
       [ 0, 27],
       [ 0, 28],
       [ 0, 29],
       [ 0, 30],
       [ 0, 31],
       [ 0, 32]], dtype=int64)

coords returns an array, which can be converted to a list:

coords_list = coords.tolist()
# return first 10 elements of the list:
>>> coords_list[:10]
[[0, 23], [0, 24], [0, 25], [0, 26], [0, 27], [0, 28], [0, 29], [0, 30], [0, 31], [0, 32]]

Update:

Based on the comment made by fmw42, if you are only looking for coordinates where difference between pixel intensities is less than or greater than a certain value (say 10); you could do the following:

sub = cv2.absDiff(img1, img2)
np.argwhere(sub > 10)
user_68859020

If you happen to also want a quantitative similarity score between the two images, here's a method using the skimage.metrics.structural_similarity function from scikit-image which returns a score and a diff image. Since you're mainly interested in all the pixel coordinates that are different, the diff image contains the actual image differences where we can search for all (x y) pixel coordinates.

Image Similarity: 34.3664%

Coordinates in Numpy form (you can convert to a list)

[[  0  20]
 [  0  21]
 [  0  22]
 ...
 [572 521]
 [572 522]
 [572 523]]

Code

from skimage.metrics import structural_similarity
import numpy as np
import cv2

# Load images as grayscale
image1 = cv2.imread('1.png', 0)
image2 = cv2.imread('2.png', 0)

# Compute SSIM between the two images
# diff contains the actual image differences between the two images
(score, diff) = structural_similarity(image1, image2, full=True)
diff = 255 - (diff * 255).astype("uint8")
print("Image Similarity: {:.4f}%".format(score * 100))

# Search for all pixels that are different 
# Type is <class 'numpy.ndarray'>, you can optionally convert to a list
coords = np.argwhere(diff > 0)
# coords = coords.tolist() 
print(coords)

cv2.imshow('diff', diff)
cv2.waitKey()
user_111621650
Related