Researching For Content-based Video Retrieval Algorithm

In modern days, data of video emerges every day. It has became a new problem thathow to find out video clips they need from the large number of diverse types of videodata. However, due to the large amount of video data which is complexity and has noorganizational structure, it has became one of the research hotspots to processing videoto make it easy for people to retrieve the fragments they need from the complex videos.For this kind of situation, after a in-depth study on traditional video retrieval algorithms,this paper proposes a shot boundary detection algorithm based on backing N/2combining with space-time slice and a key frame extraction algorithm based onswallowing cluster.The algorithm of backing N/2combining with space-time slice compares directlythe frames at a distance of N. Under the premise of using scale invariance and rotationinvariance of color feature, the algorithm extracts the space-time slice and overcomesthe disadvantage that color feature is not sensitive to spatial location. The simulation ontraditional shot boundary detection algorithms and the newly proposed algorithm bymatlab shows that, compared with traditional algorithms, the new algorithm has loweralgorithm complexity with a small fluctuation of call ratio and precision ratio.The algorithm of swallowing cluster refers to the clustering algorithm of DBSCAN,K-means, and ant colony. It regards the video frames as swallowing objects, and thenextracts video key frames through swallowing between each other. The simulation ontraditional shot boundary detection algorithms and the newly proposed algorithm bymatlab shows that, compared with traditional algorithms, the recall ratio and precisionratio have a certain degree of improvement and the algorithm has a certain applicationvalue.