Study On Video Summary Method Based On Deep Learning

As an important product of the development of contemporary multimedia information technology,video summarization technology is due to the widespread popularity of smart camera devices,and the research of video summarization technology has been more in-depth.Video summary technology mainly refers to the selection of key frame subsets or shots that best represent the original video information from the original video.In other words,the selection and extraction of key frames in video summary technology is its core work.The traditional summary algorithm Considering more the color information of the image,it is not ideal to rely on a single color feature for key frame selection.The development of deep learning provides a variety of options for solving video summary technical problems.This article mainly studies the video feature fusion and reinforcement learning mechanism.The quality of the video summary has been improved.The main work of this thesis is as follows:(1)This thesis expounds the research significance of video abstract research background and current memory display application.This paper introduces the basic framework and main summarization process of video summarization technology,analyzes the theoretical basis of traditional methods and deep learning methods,and analyzes their corresponding methods.(2)In view of the advantages of deep learning methods in data processing speed and high-latitude feature extraction,an improved feature fusion video summary method is proposed.First,the feature frames of the video are extracted by the encoder to obtain the original video features.On this basis,from the perspective of differential features and local enhancement features,then the characteristics of the decoder BRNN's dependence on long time series are used,and Through the score fusion mechanism,select key frames.Experimental results show that this method can improve the extraction effect of key frames and improve the evaluation index of video summary quality to a certain extent.(3)Thanks to the development of deep network,a method of video summarization based on feature fusion and adding reinforcement learning mechanism is proposed.Changes were made in the abstract model,and the value of the importance and diversity of key frames was summed and used as the action value function in the reinforcement learning mechanism.Through the continuous learning cycle between thevideo digest network and the reinforcement learning part,the quality of the extracted key frames is improved,and the selection and extraction of the key frames is the core of the research on video digest technology,so this method proves the feasibility and effectiveness of the reinforcement learning mechanism in the summarization technology.