文摘
This paper presents a method of near-duplicate video detection based on representative-sequence. Firstly, the video is divided into different scenes according to the variance of the Chi-squared color histogram and Grayscale OM (Ordering Measure), and then the frames of a single scene are preprocessed by decreasing the frame rate to a particular rate and detecting the occurrence of black-sides. For each scene, every certain number of frames are merged into one TIRI frame, and then weighted grayscale mutual information is calculated to select the representative frames from the TIRI frames. Next, several features are extracted from the representative frames for matching, including color, edge and grayscale features. Finally, whether or not the video is a near-duplicate is determined by the match degree of the characteristics. Experimental results show that the algorithm presented by this paper possesses good robustness and detection capabilities.