Cleans similar or duplicate images from folder using perceptual image hashing with live preview
- Reducing size of each image to NxN pixels (set by user)
- Converting to grayscale
- Calculating the mean of all color values
- Setting the color of each pixel to black if the NxN matrix value is above average, and to white otherwise
- Hamming distance: comparing hashes with each other and count the number of different pixels
- The image is considered similar and moved from the folder if hamming distance is less than a threshold set by user
Tested with Python 3.8.2, OpenCV 4.3.0, Ubuntu 20.04 LTS
Folder input_images contains 11 objects in 186 image variations created by YOLO neural detector from video
This example will leave all unique 11 objects in the folder, the rest will be moved to similar_images:
python image_hash.py -i images/input_images -o images/similar_images -s 16 -t 33
Args:
-i: input path to scan
-o: output path to move similar images
-s: hash size (default=16)
-t: threshold percent value for detection from 0 to 100, where 0 is absolute similarity of hashes (only nearly identical images will be moved)
