Research On Video Segmentation And Summarization Algorithm Based On Key Frame

With the progress of the Internet era,more and more video and audio content are displayed on various network platforms.The short videos uploaded by users,the video content released by the media,the short films produced by various institutions,and the immersive films applied to VR devices are increasing day by day.Massive film and television content enriches the entertainment methods of today's Internet users,but also makes users spend a lot of time looking for videos of interest.As a result,video shot segmentation and summarization algorithms emerged as an automated video processing technology.The video segmentation algorithm can divide the video into independent parts according to the shot and content,and the video summary algorithm can extract the most representative key frame from a video as the summary result.Saving the summary results as video data can greatly reduce the physical space used for video storage,and can also greatly compress the video time,reducing the time consumed by searching and watching the video.Therefore,for the current international mainstream video data sets,I plan to carry out video segmentation and summarization related algorithm research work,mainly focusing on the key frames of the video to improve the video segmentation processing algorithm and the static video based on the importance index of the key frames Summary of the improvement of the algorithm.The main work is as follows:1.Video segmentation processing can be divided into the boundary check algorithm research of abrupt shots and the boundary check algorithm research of gradual shots.This paper proposes a method that combines image content similarity features and deep residual networks to achieve video shot boundary segmentation,so that the video segmentation algorithm can not only target sudden shot video files,such as news clips,film and television shots,and other video files that have been edited.,It can also target gradual lens video files,such as self-media short video one-shot video files of mobile phone cameras and fixed-camera video files generated by security traffic surveillance cameras.2.The result of the static video summary is one or several independent pictures.This article will combine the various importance indicators of key frames to comprehensively calculate the importance of the image frame,and extract the image frame that best represents the video content as the summary result.The algorithm uses visual features,local features,image memorable index and image entropy as the basis for evaluating the importance of key frames,visual features as a measure of visual intuitive experience,local features are used as a measure of image detail information,image memory index is used as a high-level semantic feature to represent the result of human perception,and image information entropy is used as a measure of image content richness.These features work together in the summary algorithm to help us conform to human consciousness and experience and have representative static video summary content.