Content-based video similarity model

Abstract

The most commonly used method for content-based video retrieval is query by example. But the definition of video similarity brings great obstacle to further research. This paper puts forward a new approach to solve the difficulty. Firstly, it advances centroid feature vector of shot in order to reduce the storage of video database. Secondly, considering all the factors existing in human vision perception, it introduces a new comparison algorithm based on multi-granularity of video structure, which has great flexibility. Thirdly, after getting the similar video set, we take a brand-new method of feedback to adjust weight based on video similarity model. In this way, retrieval result can be optimized greatly.