Unified Hasher

This library aims to streamline usage of different perceptual image hashes.

Installation

pip

pip install unihasher

Details

The library provides the following methods for comparing the similarity of two image hashes:

1. Individual Hash - The verdict for whether an image was good or modified from a bad one was determined solely from a single hash algorithm.

2. Majority Decision - The similarity values for all four hashing algorithms were compared separately, and the final verdict was the verdict of the majority of the hash algorithms. In the case of a tie, the verdict of the best performing hash from Approach 1 was taken.

3. Decision Tree - The similarity values for a combination of all four hashing algorithms were considered by passing the values through a decision tree.

The hashing algorithms implemented are:

dhash, phash, whash from imagehash library

nmfhash adapted from Robust Perceptual Image Hashing Based on Ring Partition and NMF (Tang et al.)

For more details, please refer to our paper.

Made by: Akshara Mantha, Peng Ruijia, Tan Siying

Usage

Please refer to unihasher_demo/unihasher_usage.py for details on how you may use the library.