Stereoscopic Video Clip Matching Algorithm Based on Incidence Matrix of Similar Key Frames

Abstract

Clip matching is the key of content-based video retrieval. In order to better realize the content-based video retrieval and improve the accuracy and efficiency, a stereoscopic video clip matching algorithm based on incidence matrix of similar key frames is proposed by integrating depth information in feature extraction. In the algorithm, the stereoscopic videos are divided into sub-sequences through shot and sub-shot segmentation, and the key frames and features are extracted. With the features of key frames, this paper proposed the visual similarity based on the incidence matrix of similar key frames, order similarity based on the maximum common sub-sequence of linearly fitting, semantic similarity based on the cross similarity of similar key frames. And then the linear expression of similar matching is built so as to realize the similarity calculation and effective matching of stereoscopic video clips. In the experiment of clip matching, the mean average recall (MAR) is increased by 3.69 and 0.29%, the mean average precision (MAP) is increased by 10.17 and 8.10%, and the mean average matching time (MAMT) is decreased by 44.35 and 38.60%, respectively, compared with the existing typical stereoscopic video retrieval algorithm SVR and the video clip matching algorithm VTD-LFF. The experimental results show that the proposed algorithm can realize stereoscopic video clip matching more accurately and efficiently.