| With the rapid development of multimedia technology and Internet, more and morevideo materials could be available. In video processing and analysis, how to save andmanage videos effectively, how to retrieval the video you want quickly and accuratelyin a gigantic video database have become a crucial problem of the multimediatechnology research.Video scene segmentation is an important and hot technology in video analysis.According to the related contents and temporal adjacency of the shots, the similar shotsare divided into the same scene. Therefore, a complete video can be segmented into acertain number of logical story units that have semantic meaning, which is important forvideo research.In order to solve the problem of the semantic gap between low-level features andhigh-level semantics for video scene segmentation, on the basis of shot key frameextraction, a semantic concept vector is introduced and a semantic-based sceneconstruction method is proposed. A multi-class support vector machine (SVM) is usedto classify the video key frames of scenery documentary. The color, texture and shapefeatures from the video key frames of scenery documentary are combined to classify thekey frame in SVM in order to obtain the semantics of the keyframe. The performancesof SVMs with difference kernel functions are analyzed and compared. and the resultsshow that the support vector machines with RBF kernel are more accurate.In this paper, a method for shot overlapping link algorithm based on semantic isproposed. First, the support vector machine is used to extract the semantics of keyframes. Then, the semantic concept vectors of key frames are formed. The shot keyframes are clustered by the same semantic concept vectors to construct the video scene.The method can avoid the problem of threshold set of the general scene segmentationalgorithm. The experimental results show that the proposed method can improve thecorrect rate of scene segmentation and has a certain universality for the same type ofvideo. |
PDF Full Text Request