Research On Video Semantic Concept Analysis Based On Group Sparse Representation

In recent times,there have been advancements in the growth of the Internet technology.As a result,there are numerous video resources scattered in diverse repositories(sources)in an ad hoc fashion.Retrieval of videos from these sources with conventional video management tools has some challenges in meeting the needs of contemporary users.This challenge informed researchers to pay more attention to video semantic analysis.It is difficult to close the gap between low-level visual features of video and semantic understanding.How to extract easily semantic concept for easy understanding from video data is still an urgent problem to be solved in the study of video semantic analysis.Again,it must be emphasized that the existing techniques implemented based on sparse representation and dictionary learning do not take into full consideration the locality structure of video semantic data containing more discriminative information,which is very crucial for classification.In addition,similar coding outcomes were not being realized from video features belonging to the same video category.Furthermore,videos are nonlinearly structured in nature and the existing techniques fail to fully capture the nonlinear discriminative information hidden in the structure of video semantic data,which is essential for efficient classification results.In this dissertation,the background of the study and the application areas of video analysis with sparse representation were firstly outlined.After which the research status is discussed briefly and finally the preprocessing aspect of video semantic analysis is introduced.This includes video shot segmentation,video key-frame extraction,video feature extraction and feature fusion.Throughout this current research,a significant progress has been made in its application with video semantic analysis.This was achieved by putting forward group sparsity based locality-sensitive dictionary learning for video semantic analysis,kernel locality-sensitive discriminative sparse representation for video semantic analysis and a kernel weighted KNN classification method based on kernel localitysensitive discriminative sparse representation for video semantic analysis.The main contents are as follow:1)A locality-sensitive discriminant group sparse representation method is put forward.In this study video samples of the same category were encoded as similar sparse codes in the process of video semantic analysis based on sparse representation.In other words,sparse coefficients of two video features from the same category were assumed to be similar,so as to enhance the power of discrimination of sparse representation features.In the proposed method,discriminant loss function based on sparse coefficient is introduced into the locality-sensitive sparse representation.An optimization dictionary is generated with the constraint.The sparse features have both small withinclass scatter and large between-class scatter,so as to build the discriminant sparse model.The experimental results show that this method significantly enhances the power of discrimination of sparse representation features,and consequently improves the accuracy of video semantic analysis in all metrics compared with other state-of-the-art methods.2)Mapping data to high-dimensional space can improve the discriminability of data.A discriminative classification algorithm based on kernel locality-sensitive sparse representation with group sparsity is proposed.This was done by integrating 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 kernel mapping.The 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 group sparse representation coefficients is added to the objective function to make the sparse representation more discriminative.Experimental results show that the proposed method further improves the accuracy of video semantic concept detection in all metrics compared with the state-of-the-art baseline approaches.3)A weighted KNN classification method based on the reconstruction error of the kernel locality-sensitive discriminant sparse representation is proposed.The reconstruction error classification methods based on sparse coefficient does not consider discrimination between video samples sparse coefficients,and this consequently affects the accuracy of video semantic classification.This section puts forward loss function that integrates reconstruction error and discrimination.The proposed weighted KNN classification approach is used to calculate the loss function value between the test sample and training samples from each class according to the loss function criterion,and then vote on statistical results.Finally,we modify the vote results combined with the weight coefficient of each class and determine the video semantic concept.The experimental results show that this method effectively improves the accuracy of video semantic analysis and a slightly high computational time used in the semantic classification compared with some baseline approaches.