Implementing a Perceptual Hash Algorithm for UI Image Similarity in Android Automation

In the 360 testing project, it is necessary to compare the similarity of two UI screens and perform corresponding actions based on the result.

A naive double‑loop comparison has high time complexity; introducing a hash algorithm before traversal can significantly reduce execution time.

The perceptual hash algorithm creates a fingerprint string for each image; comparing these strings yields a similarity measure, where a smaller difference indicates higher similarity.

Implementation steps include:

Resize the image to 8×8 pixels (64 total pixels).

Convert the RGB image to grayscale using the weighted formula Gray = 0.299 R + 0.587 G + 0.114 B.

Calculate the average gray value of the 64 pixels.

Generate a 64‑bit binary value by setting each bit to 1 if the pixel’s gray value is greater than or equal to the average, otherwise 0.

The resulting binary string is the image’s hash (fingerprint).

Compare the two hashes; a diffNum < 5 indicates the images are very similar, while diffNum > 10 means they are completely different.

By capturing screenshots with UIAutomator before and after UI changes, the similarity score can be used to decide whether further actions are needed.

Reference implementation can be found at: https://github.com/gavinliu/SimilarPhoto/blob/master/app/src/main/java/cn/gavinliu/similar/photo/SimilarPhoto.java