A Video Retrieval Algorithm Based on Ensemble Similarity
This paper proposed an ensemble similarity based method for video retrieval. An ensemble similarity is used to calibrate the similarity between user given query video clip and each video clip in the database: a clip can be treated as an ensemble which consists of a sequence of multiple kev frames. By kernel method, in a high dimension space the feature vector represented frames can be assumed to distribute a Gaussian model. Then probabilistic distance between two Gaussians is computed as the similarity value between two video clips. Then video clips in database with the highest similarity are output and submitted to the user. To improve the speed efficiency, an improved algorithm of Chernoff distance and KL divergence is also proposed. The experimental results indicate that the proposed approach achieves superior performance than some existing methods.
video retrieval ensemble similarity probabilistic distacne kernel method
Li Deng Li-Zuo Jin
Key Laboratory of Power Station Automation Shanghai University Shanghai, China Department of Automatic Control Engineering Southeast University Nanjing, China
国际会议
厦门
英文
638-642
2010-10-29（万方平台首次上网日期，不代表论文的发表时间）