Discriminant Self-adapted Locality-sensitive Sparse Representation For Video Semantic Analysis

In recent years,with the explosive growth of video data,the research of video semantic analysis is becoming a research hot spot.In video semantic analysis,the diversity of scene for the same semantic content in video always exists objectively.Even the different scenes in the same video semantic category may lead to different video features.How to improve the accuracy of video semantic concept detection by tolerating the diversity of the scene is always a problem to be solved.In this thesis,the background,significance,and current situation of video semantic analysis are firstly described.And then,the related sparse representation algorithms and kernel sparse representation theory are briefly described.Combining the advantages of sparse representation algorithms and the development requirement of video semantic analysis,this thesis proposes discriminative self-adapted locality-sensitive sparse representation dictionary learning method,a classification method based on discriminative self-adapted locality-sensitive kernel sparse representation,and a prototype system for video semantic detection based on the discriminative self-adapted locality-sensitive sparse representation dictionary learning method.The main works of this thesis are as follows:1)In the process of sparse representation of video features,dictionary learning method can be used to get potential relationships among different video semantic features.A discriminative self-adaptive locality-sensitive dictionary learning method(DSALSDL)is proposed in this thesis.The method constructs an adaptive local adapter and adds it to the learning process of sparse representation dictionary,thus obtaining the structural information of video data.In addition,in self-adaptive locality-sensitive sparse representation,the discrimination loss function with class discriminability is introduced,which makes the sparse representation feature have a stronger aggregation correlation within the intra-class and have a greater dispersion between the inter-classes of the video feature samples.The algorithm based on discriminative self-adaptive locality-sensitive sparse representation is used to detect the semantic concept of video.The experimental results show that this method can improve the accuracy of video semantic concept detection quickly and effectively.2)Mapping data to high-dimensional space can improve the discriminability of data.A classification algorithm based on discriminative self-adapted locality-sensitive kernel sparse representation is proposed.By combining the kernel function with the sparse representation,the video data sample and the sparse representation dictionary are mapped to the high dimensional kernel feature space by the kernel mapping.The self-adaptive local adaptor is constructed as the constraint term of the kernel sparse representation by using the dictionary atoms and training samples in the high dimensional space to optimize the sparse representation coefficient.At the same time,a discriminative loss function based on sparse representation coefficients is added to the objective function to make the sparse representation more discriminative.Experimental results show that the proposed method improves the accuracy of video semantic concept detection.3)A prototype system of video semantic concept detection based on adaptive local sensitive discriminable sparse representation dictionary learning algorithm is designed and implemented.The system uses the hybrid programming technology of VC++ and MATLAB,and uses object-oriented design method to design and implement three modules,model training,video playing and video semantic detection,and integrates the realization of video semantic concept detection prototype system.The system operation shows the availability of the system.