The Technology And Implementation Of Video Summarization Based On Content Analysis

With the development of science and technology and the popularization of mobile camera equipment,more and more people can shoot and publish videos at anytime and anywhere,resulting in the generation of large video data.It has been a challenge that how to efficiently locate the desired information in mass video.video summarization is a skill that is able to generate a brief video which automatically captures the important information and interesting events in video content,which helps to mitigate this problem.Traditional video summarization algorithms are generally unsupervised learning methods.By defining the objective function to select the important content,these algorithms produce a large number of complex computations,which are inefficient.Static video summaries often fail to retain the original motion content and audio information,resulting in incomplete information acquisition.Aiming at some existing problems in video summarization algorithms,based on the analysis of video content,this paper proposes a new solution.This paper is mainly divided into the following three parts:(1)Dynamic video summarization based on consistent clip generation.Reasonable segmentation of video clips plays an important role in video summarization algorithm.Most traditional video segmentation methods are mainly aimed at the video edited by users.Utilizing clip similarity score and local similarity score,the algorithm designs a new clip segmentation method,which divides the original video into multiple consistent clips through three steps of cutting,merging and adjusting.Then,using a set of features,the algorithm estimates the importance of consistent clip content to produce high quality video summary.(2)Dynamic video summarization based on long short-term memory(LSTM).Utilizing supervised learning technology,the algorithm designs a model structure,which consisting of bidirectional LSTM layer,multi-layer perceptron(MLP)and determinantal point process(DPP).This algorithm can simultaneously model the importance,diversity and logic of video frames in video content.By training the model,this method is able to generate a good video summary.(3)The video summarization algorithm proposed in this paper is designed and implemented.The video summarization demo system can input any video specified by the user and play the video,pause the play,jump to the specified content and so on.The system has the function of video summarization.