Creating Video Summarization From Emotion Perspective

With rapid development of Internet and mobile devices,video data is increasing sharply.Thus,users usually cannot efficiently find out particular videos they intend to watch.To address this problem,the technology of video summarization has attracted great attention gradually from researchers.However,most of existing video summarization algorithms attempted to extract summarization based on video content,while the emotion information is not taken into account.In this thesis,we focus mainly on emotion based video summarization problems.In particular,two kinds of video summarization problems are exploited,i.e.,dynamic video summarization and video cover extraction.The main contribution of this thesis can be summarized as follows:(1)Based on regression model,a new dynamic video summarization algorithm is proposed from emotion perspective.First,we construct a new video dataset and annotate it manually by employing subjects.Then,we propose a new emotion quantification approach and a new emotion feature representation scheme,which extract emotion feature for the dataset.Based on above foundation,a regression model can be trained,which can exactly predict the emotion intensity of each frame in a video.Finally through this regression model and optimization algorithm,a more representative video summarization can be generated.From both of objective and subject perspectives,experimental results illustrate the effectiveness and superiority of the proposed algorithm.(2)Based on emotion information,a new algorithm for video cover selection is proposed.First,we consider emotion intensity,image blurring degree and consistency as criteria to select appropriate frame as video cover.After that,edge and contrast enhancement are utilized to improve the quality of video cover.Experiment results reveal that the proposed method is more effective and the generated cover is more expressive and attractive.