Application Of Visual Evaluation Association Modeling Based On Chaotic Pseudo-random Multi-layer Compressed Sensing Privacy Protection Keyframe Extraction

In today’s society,with the rapid development of information industries such as artificial intelligence on a large scale,people are exposed to big data all the time in their lives.In today’s social context where people are paying more and more attention to public safety issues,the video surveillance system that has been relied on for a long time has caused more and more frequent privacy leaks,which will not only cause certain troubles to people’s lives,but also It will bring huge losses to the society.On the other hand,the rapid development of video surveillance technology will also generate a large amount of video image data.The storage and processing of these data will bring huge resource consumption,and will also affect the development of society to a certain extent.Therefore,it is a very meaningful research direction to reduce the cost of data storage while ensuring that the information is basically not lost.The traditional video privacy protection method is to protect the privacy information in the process of sampling,compression and transmission,so as to avoid the privacy leakage caused by the video data falling into the hands of others.However,the above methods do not take into account the necessary conditions for the development of subsequent intelligent video applications,nor do they take into account the problem that more time and computation are needed for the subsequent processing of video data after being processed by these methods.Therefore,this thesis adopts a new video coding technology,which does not need to encrypt,decrypt,reconstruct and other tedious ways like the traditional way,but realizes the privacy protection on the visual level in the process of relevant intelligent application of video data.In order to solve the problem of privacy information disclosure in video surveillance data,and at the same time take into account the high storage cost and high labor consumption in retrieval process caused by massive video data generated by widely distributed intelligent surveillance devices,the innovative research part of this thesis mainly includes the following aspects:(1)Using compressive sensing theory,a new improved observation matrix is proposed by combining the observation matrix with chaotic pseudo-random technology.Using the above technology to perform multi-layer visual privacy protection encoding processing on the video,it can achieve privacy protection at the visual level,and at the same time reduce the loss of internal feature information as much as possible,so as to ensure the normal development of the intelligent application of subsequent key frame extraction.corresponding guarantee.(2)In order to objectively and effectively measure the quality of video privacy protection,and to avoid the influence of uncertain factors such as external equipment and subjective factors on the measurement of video privacy protection quality,this thesis proposes a LBP feature algorithm based on a unified visual saliency model based on graph theory.and a contrast feature extraction operator based on the statistical mean of the asymmetric improved a filter,an improved two-feature fusion video privacy protection quality evaluation algorithm is designed.The evaluation results of the algorithm can more effectively describe the video privacy protection quality.(3)In view of the problems caused by the storage of video data and excessive privacy protection,this thesis proposes a two-layer unsupervised clustering key extraction algorithm based on affinity propagation clustering and sparse subspace clustering,which can achieve effective and concise reflect the video content.At the same time,this thesis also proposes the CF value to evaluate the performance index of the key frame extraction algorithm.In the case of lack of information expression between the video after privacy protection encoding and the original video,excessive privacy protection will lead to the performance of the key frame extraction algorithm becoming meaningless.Therefore,this thesis establishes a video privacy protection quality score and A correlation model of the performance of the keyframe extraction algorithm to balance the relationship between the two.The video privacy protection score is used to guide the performance of the key frame extraction algorithm to avoid excessive privacy protection problems.