Efficient and Effective State-based Framework for News Video Retrival

In this paper, an efficient and effective framework is proposed for news video retrieval. Firstly, the 64-dimensional colour histogram is extracted as the feature vector. Then the pair quantizer is adopted to transfer the news video retrieval problem into multi-dimensional string matching problem, which conduces to the efficiency to the framework. Secondly, a new measurement named ‘optimal temporal common subsequence’, which distinguishes the difference caused by rich temporal characteristics including temporal order, temporal duration and temporal gap, is designed to match state-sequence, followed by the point & interval-based formal characterization of time-series and state-sequences. Thirdly, we tested the proposed measurement on news video retrieval. The performance shows the proposed algorithm is more effective for news video retrieval.