A Video Semantic Analysis Method Based on Kernel Discriminative Sparse Representation and Weighted KNN

摘要

To improve the classification performance of sparse representation features, a method of video semantic analysis based on kernel discriminative sparse representation and weighted KNN is proposed in this paper. A discriminative model is built by introducing kernel category function to KSVD dictionary optimization algorithm, mapping the sparse representation features into high-dimensional space. Then the optimal dictionary is generated and applied to compute the sparse representation coefficients of video features. Finally, the video semantic analysis is made by means of weighed KNN method based on optimization sparse representation. Before the video semantic analysis, genetic algorithm is used to get global optimal features and reduce the dimension. Furthermore, the kernel function is introduced to establish discrimination about sparse representation features and the classification vote result is weighed, the purpose of which is to improve the accuracy and rationality of video semantic analysis. The experimental results show that the proposed method significantly improves the discrimination of sparse representation features and is 22.33% higher in accuracy compared with the traditional SVM method based on KSVD. The method is suitable for the classification of video features with nonlinear relationship, tolerating not only the noise but also interference problems in video shot.