考虑到图像遮挡后部分局部信息属性改变，在利用最小残差判决函数分类时，各类残差可能因较接近而导致分类错误。针对此问题，从分类器判决函数出发，提出基于稀疏系数累积的局部-全局加权融合的稀疏表示遮挡人脸识别算法。该算法主要利用各类稀疏表示系数累积作为判决函数，使用Borda投票机制进行分类。利用系数累积进行全局分类，然后对局部各块分类，考虑到子块作用不同，利用稀疏度和残差两个参数表示其可信度权重，最后将全局和局部融合Borda投票，统计各类投票总数，实现分类。在公用数据库进行实验，结果表明该算法具有较好的有效性和鲁棒性。%In the occlusion face recognition, some covered parts change the property of local information. It may lead to a wrong classification using the minimum residual as a decision function for sparse representation classification when the residual is approximate. In this case, proceeding from the decision rule of the classifier, the algorithm of sparse representation with weighted fusion of local based non-minimum square error and global is proposed for face recognition. The accumulation of each class of coefficient is mainly used as the decision function and the Borda votes system is introduced for sparse representation classification. Firstly, the sparse coefficient accumulation of each class is calculated for global classification. Then, for the local information, the subblocks coefficient accumulation is used to classify. Considering the different effects of subblocks, the sparsity and residual are utilized to jointly express the weight of credibility. Finally, the global and local blocks are combined to Borda vote for the final classification. The experimental results on public available database demonstrate that the proposed algorithm has good effectiveness and robustness.
展开▼
