Visual Shielding Compressed Sensing Coding And Double-layer Affinity Propagation For Video Summaries Generation

Nowadays,with the rapid development of Internet technology,people learn,entertain or watch news through video.Video data is an indispensable part of our daily life.However,due to the large amount of video data,how to quickly and efficiently obtain video content information is a difficult problem that must be explored.Researchers' years of exploration have clearly proved that video summary technology is an effective way to reduce the amount of video data and accelerate the speed of obtaining information in videos.Traditional video summary technology has high requirements for video resolution.For videos that involve privacy protection,most existing literatures do not have specific solutions.This paper proposes a new key frame extraction algorithm for video data in the visual shielding domain.(1)Visual Shielding Compressed Sensing(VSCS)algorithm is proposed.Compressed Sensing(CS)technology is used to reduce the dimensionality of the original video to obtain the compressed video.Then,the video has the effect of visual shielding(to protect the privacy of video content),and the amount of data is greatly reduced.The key premise for realizing third-party real-time supervision of videos that require privacy protection is to ensure that the specific content of the video is not visible.If video data is understood as a set of time-series images,the protection of video content should undoubtedly take images as the processing object.The method of video processing in this paper is different from simple data encryption.The focus should be on the visual shielding of video frames,that is,the data to be processed for supervision can be below the lower limit of human visual perception while the content is fidelity.In addition,the introduction of CS sampling is not only to complete the visual shielding,but also to use its high-fidelity characteristics of the timing relationship and frame content to solve the key frame extraction problem of the reduced-dimensional video.(2)Double Affinity Propagation(DAP)algorithm is proposed.We enter the videos with privacy features into pre-extract key frame system,initially extract the PHOG features of each frame,use AP clustering to obtain video summary sequence of the first stage.Then input the video summary of the first stage into the secondary extraction key frame system.For each input key frame,the PHOG-Hist fusion feature is extracted and AP clustering is used again.In the second step of clustering,the frame number of the video used for clustering has been greatly reduced,so extracting the fusion features and clustering does not take much time to obtain the final output video summary sequence.(3)A method for extracting video features in stages is proposed.The method of uniformly extracting fusion features for all frames included in the video requires extracting multiple features of each frame in the video at the same time,which is relatively computationally intensive.Moreover,the experiments in this paper verify that no matter whether it is for the original video or compressed video,the method of fetching all features at once and then clustering multiple times is not the optimal solution.In this paper,a single feature is extracted first for clustering.Then the AP algorithm is used for the second time to cluster the fusion features extracted from the key frames obtained in the first stage.Compared with other advanced methods,the proposed method is efficient and accurate.