H. Yang, X. Hou, and L. Guan (PRC)
Local self-similarity, feature descriptor, rate-distortion trade-off
The local feature description is usually the first step of image or video processing. In this paper we propose a new local self-similarity method based on key local patches or temporal frames detection, which give the representation of the whole visual entities (images or videos). For a given image or video sequence we compute distances between representations for all pairs of spatial patches or time-frames and store results in a Self Similarity Matrix (SSM) defined as “the local feature descriptor”. In order to filter out non-informative descriptors, an automatic key entities (patches or frames) selection method according to a rate-distortion trade-off of SSM is presented. The method evaluates the globally optimal set of patches or frames to represent the entire object or sequence without requiring pre-segmentation. Comparative evaluation of local key feature descriptor with previous methods demonstrates improved performance.
Important Links:
Go Back
