Research On Key Issues Of Content-based Video Summerization

With the rapid development of the Internet,the amount of multimedia video data is growing rapidly,and the content structure of video is becoming more and more complex.How to effectively store and manage these video data is particularly important.Video summary is a summary of complete video,which compresses the video length while retaining the original information.An effective video abstraction technology can greatly reduce the storage space of video and improve the efficiency of video analysis.This paper mainly studies shot segmentation and key frame extraction in video abstraction technology.The main work is as follows:1.An abrupt shot boundary detection algorithm based on color and depth information is proposed.Aiming at the low robustness of single feature detection results of existing mutation shot detection algorithms,an abrupt shot boundary detection algorithm based on color and deep information is proposed by analyzing the relationship between color features and CNN extracting depth features and shot changes.The feature synchronization extraction model realizes the detection of potential mutation frames,and further realizes the final detection of mutation frames through flash detection module and large object detection module,so as to improve the detection accuracy.2.In view of the limitation of threshold selection and the high complexity of gradual shot detection method based on low level feature modeling,a gradual shot detection algorithm based on single-stage C3 D network is proposed by analyzing the correlation between abrupt shot and gradient shot.On the basis of the detection results of abrupt shot,a gradual shot detection model is constructed to realize gradual shot detection.A gradual shot detection algorithm based on multi-stage C3 D network is proposed to improve the localizing accuracy of the algorithm.3.A key frame extraction algorithm based on multi-feature clustering is proposed.Aiming at the problem that single color or texture feature analysis method is difficult to detect redundant frames due to the change of view angle in the process of shot moving,considering that ASIFT features can imitate different angles of camera and have strong affine and scale invariance,a key frame extraction algorithm based on multi-feature clustering is proposed.A two-stage detection model is constructed to extract key frames accurately.