Video Semantic Analysis Based On Self-adaptive Local Sensitive Discriminant Group Sparse Representation

With the high-speed development of multimedia and Internet technology,video information in the Internet is increasing rapidly.How to quickly and accurately find the video information which can meet the needs of users from massive data has become a hot research topic for many scholars.However,due to the rocketing data size and the rapid inflation of data information,the traditonal video retrieval methods have been unable to meet the growing needs of users.Because it is difficult to cross the gap between low-level features of video and high-level semantic concept,how to cross the gap and extract the high-level semantic concept which is easy to understand has become one of the hot issues to be resolved in the study of video semantic analysis.Firstly,thesis briefly introduces the research background,research significance and situation of video semantic analysis.Then thesis detailedly introduces sparse representation theory and it's existing applications in the video semantic analysis.According to the further improvement of some sparse representation methods and the demands for video semantic analysis,thesis proposes a locality-sensitive discriminant group sparse representation method,self-adaptive locality sensitive discriminant group sparse representation method,and designs a prototype system of video semantic analysis based on the above two methods.The main research contents are as follow:(1)A local sensitive discriminant group sparse representation method for video semantic analysis is proposed.In this thesis,it analyzes the locality-sensitive sparse representation model,and considers the selection of samples variables with the similarity of video data in the same category,and then integrates group sparse model and discriminant function,which develop a locality sensitive and discriminant group sparse representation method.This proposed method realizes variable group selection ability and renders the sparse features to satisfy the fisher criteria.Compared with other methods,the experimental results show that the proposed method can improve the accuracy of video semantic analysis more effectively.(2)In order to further enhance the ability of keeping data local structure andpotential information for the dictionary,this thesis presents a self-adaptive locality-sensitive discriminant group sparse representation method.The presented method considers the relationship of linear expression and reconstitution between dictionary atoms,which builds a self adaptive local adaptor.Then the designed self adaptive local adaptor is used to guide the sparse coding process gradually.This method iterates the stages of establishment of self adaptive local adaptor,sparse coding and dictionary updating,so as to obtain the final optimized dictionary.Compared with other methods,the experimental results show that the proposed method can improve the accuracy of video semantic analysis further effectively.(3)This thesis adopts the object-oriented design philosophy and applies the above proposed methods,which designs a video semantic detection prototype system based on self-adaptive locality-sensitive discriminant group sparse representation method.The system mianly contains three modules: video preprocessing,model training and video semantic detection,which designs a friendly interface with easy accessibility.The final running test results show that the methods proposed in this thesis have a certain availability.