Image Hash using Neural Networks

Abstract

Hash functions have been used to generate hash codes for data authentication. Traditionally these functions are generated using byte oriented algorithms like MD5 and others. In our paper we propose a new method of generating hash code for images using neural networks. Three sample images namely, fingerprint, lena and football image have been considered and their hash values calculated using two neural network structures namely, 1) structure without feedback 2) structure with feedback. The original images are then subjected to bit modification,Gaussian noise and rotational noise. The hash values are recalculated for the modified images. Sensitivity and hit collision are calculated and are found to be comparable with that of MD5 algorithm. General Terms Neural Networks, Cryptography, Security