Fragile and semi-fragile image authentication based on image self-similarity

We propose a new approach for fragile and semi-fragile image authentication based on image self-similarity. The authentication signature is extracted from image representation using generalized finite automata which represents the image in terms of its perceptual self-similarity. This self-similarity is more likely to be damaged due to illegitimate manipulations rather than compression algorithms such as JPEG. The proposed technique is also capable of changing the degree of fragility by setting the value of the image self-similarity parameter α. Furthermore, our approach features an extremely compact signature length whilst maintaining a rather simple implementation. Experimental results are presented for well known images to support these claims.