How to compare two images in OpenCV Python?

Image comparison plays a critical role in various computer vision applications, such as object tracking, face recognition, and image search. Python and OpenCV, an open-source computer vision library, offer several methods to compare two images.

This article focuses on the techniques used in OpenCV Python for image comparison and how to implement them step-by-step.

Pixel by pixel comparison

One of the most straightforward methods to compare two images is by examining each pixel value and comparing it with the corresponding pixel value in the other image. If the two images are identical, pixel by pixel comparison would result in the same value for every pixel.

Here is a basic code snippet for pixel by pixel comparison:

import cv2

# Load images
img1 = cv2.imread('image1.jpg')
img2 = cv2.imread('image2.jpg')

if img1.shape == img2.shape:
    # Calculate difference
    difference = cv2.subtract(img1, img2)
    b, g, r = cv2.split(difference)

    # Check if the images are identical
    if cv2.countNonZero(b) == 0 and cv2.countNonZero(g) == 0 and cv2.countNonZero(r) == 0:
        print("The images are identical")
    else:
        print("The images are not identical")
else:
    print("The images have different sizes")

In this code, cv2.imread() loads the two images into memory, and cv2.subtract() calculates the difference between the two images. The cv2.split() function takes the difference image and returns three matrices of pixel values for each color channel.

The cv2.countNonZero() function then counts the number of non-zero pixels in each channel. If the count for each channel is 0, then the images are identical, and if any of the channels have non-zero values, the images are not identical.

Label: Python

Structural similarity comparison

Pixel by pixel comparison is not always the best way to compare two images. Structural similarity comparison, which analyses the differences in the perceptual content between two images, is a more advanced technique. The Structural Similarity Index (SSIM) is one of the most popular methods for structural similarity comparison and is commonly used to determine image quality.

Here’s how to implement the SSIM method in Python using OpenCV:

import cv2

# Load images
img1 = cv2.imread('image1.jpg')
img2 = cv2.imread('image2.jpg')

# Convert images to grayscale
gray_img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray_img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

# Calculate SSIM
SSIM = cv2.Ssim(gray_img1, gray_img2)
print("The SSIM value is:", SSIM)

The code above loads two images, converts them to grayscale, and calculates the SSIM value. This method returns a value between -1 and 1, indicating the structural similarity of the two images. -1 indicates that the images are entirely dissimilar, 0 indicates that the similarities of the two images are random, and 1 implies they are precisely identical.

Label: Python

Mean squared error comparison

Another method to compare images is using the Mean Squared Error (MSE), which is a measure of the average squared difference between the original and predicted values in a data sample. In image processing, the MSE value is calculated by finding the difference between each pixel value of the two images and squaring it. The average of these squared differences gives us the MSE value.

The code to use MSE in OpenCV is:

import cv2
import numpy as np

# Load images
img1 = cv2.imread('image1.jpg')
img2 = cv2.imread('image2.jpg')

# Calculate mean squared error
mse = np.mean((img1 - img2) ** 2)
print("The mean squared error is:", mse)

In this code, img1 and img2 are the two images we want to compare. The ** operation squares the difference between each pixel of the two images. Finally, the np.mean() method calculates the average of all squared differences, giving us the MSE value.

Label: Python

Normalized cross-correlation

Normalized Cross-Correlation (NCC) is another image-comparison technique that compares two images by calculating the correlation of the pixels. In NCC, the pixel values of the images are normalized to have zero mean and unit variance. Next, the two images’ pixels are compared by taking the dot product of the two images’ pixel values.

Here is how to implement the NCC method in Python using OpenCV:

import cv2

# Normalized cross-correlation

# Load images
img1 = cv2.imread('image1.jpg')
img2 = cv2.imread('image2.jpg')

# Convert images to grayscale
gray_img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray_img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

# Normalize images
norm_img1 = cv2.normalize(gray_img1, None, alpha=0, beta=1, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_32F)
norm_img2 = cv2.normalize(gray_img2, None, alpha=0, beta=1, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_32F)

# Calculate NCC
ncc = cv2.matchTemplate(norm_img1, norm_img2, cv2.TM_CCORR_NORMED)[0][0]
print("The NCC value is:", ncc)

The code above loads two images, converts them to grayscale, and normalizes them using the OpenCV cv2.normalize() function. We then apply the NCC method using the cv2.matchTemplate() function, which returns a correlation map. We extract the correlation value from the map and print it out.

Label: Python

Conclusion

Comparing two images is a crucial aspect of image processing and computer vision. OpenCV Python offers several options to help with image comparison, including pixel by pixel comparison, structural similarity comparison, mean squared error comparison, and normalized cross-correlation.

When choosing which method to use, consider the application’s requirements, the speed and computational complexity of the algorithms, and the sensitivity of the comparison.

By utilising OpenCV Python and implementing these methods, you’ll be able to effortlessly and effectively compare two images, making computer vision and image processing much easier to manage.