This is a Python program for detecting similar images within a given repository. It is mean to be used over time, however, a full db-clean should not be too costly (as long as you don't have too many images) so it is currently being performed on schema updates.
The current plan is to employ three main methods of detecting similarity in images. In each case, the image is scaled down and grayscaled.
- Average Hashing
- Takes the average color of the image and determines if each pixel is above or below the average color. A bit-array representation of that is is converted into a hex representation. The fastest algorithm, but the least precise since a smaller image is used in reduction.
- Perceptual Hashing
- Uses a discrete cosine transform to distill the lowest frequency values in an image to determine a fingerprint, which is then hashed. Similar to the average hash, but to an extreme. The slowest, but most precise.
- Difference Hashing
- Creates a hash based on the color gradient across an image. Each color in a row is compared to the color directly next to it, and bit values represent a positive or negative change. A faster algorithm with high precision (not as fast as average, but not as precise as perceptual -- a good in-between).
Comparisons are done between the hashes using a Hamming Distance.
- Implement hashing algorithms
- Asynchronously calculate information for images
- Asynchronously compare images
- Compare image and determine what to do when close matches are found
- Store image information in db to reduce calculation time
- Implement method to ignore similar images on future runs
- Allow user to specify tolerance
- Add image selection to randomly select images for use