Video copy detection based on improved affinity propagation

Abstract

To solve the problems in the traditional BoVW (bag of visual words) method based on the K-means clustering algorithm such as bad searching precision and low time efficiency. An improved BoVW making method based on Affinity Propagation (AP) and applying for large scale dataset is proposed. E2LSH is used to match the similar video frame and vote to completing the video copy detection, and an LAP method is introduced to make the BoVW scalable. Experimental results show that the accuracy of the proposed method is substantially improved compared to the traditional methods. Besides, it also adapts large scale datasets well.