Near-duplicate video clustering using multiple complementary video signatures

A near-duplicate video clustering algorithm based on multiple complementary video signatures is proposed in this work. We use three kinds of frame descriptors: RGB histogram, color name histogram, and ternary pattern. Then, we convert each kind of frame descriptors for a video into a video signature based on the bag-of-visual-words scheme. Consequently, we have three signatures to represent the video. These signatures are complementary to one another, since they are robust to different near-duplication types. Also, we develop a clustering technique to refine pairwise matching results and categorize near-duplicate videos. Experimental results on an extensive video dataset show that the proposed algorithm detects near-duplicate videos more effectively than conventional algorithms.