A New Local Self-similarity Descriptor Based on Structural Similarity Index

The local self-similarity descriptor is a kind of important image or video local feature description method. It is often used for detection, identification and recognition. In this paper we propose a new local self-similarity descriptor based on structural similarity (SSIM) index. It is showed in this paper that the SSIM Index give very different answers to the question of how self-similar local patches really are. For a given image we compute SSIM index distances between representations for all pairs of spatial-patches and store the results in a Self-Similarity Matrix (SSM) defined as "the local feature descriptor". This new method is easily extended to the wavelet representation of images. Comparative evaluation of local feature descriptor with previous methods demonstrates improved performance.