基于加权分块稀疏表示的步态识别

摘要

为解决步态识别中的小面积遮挡问题,结合人体不同部位的特征对个体识别的贡献度不同这一特性,提出一种基于加权分块稀疏表示的步态识别方法.对视频图像进行处理,获得目标的GEI图像;将GEI图像划分为多个区块,独立地对每个区块的图像做稀疏表示,利用一种加权方法获得图像子区块的权重;在此基础上,对测试样本进行加权稀疏表示,获得对测试目标的分类.实验在CASIA_B数据库上进行,实验结果表明,该方法对小面积遮挡有较好的鲁棒性且识别率高于其它现有分类方法.%To solve the small area occlusion problems in gait recognition,a method based on weighted block sparse representation was proposed which was combined with the characteristics that the contribution of different parts of the bodies to identification is different.Video image was processed to get gait energy image (GEI) and GEI was divided into several blocks.Block sparse representation was introduced to classify various blocks of GEI respectively and an original weighting scheme was proposed to generate the weight value.On the basis of these processing,a sparse representation weighted model was constructed to complete the gait recognition tasks.The proposed method was tested and evaluated on CASIA database (Datasct _ B).Experimental results suggest that the proposed method shows robustness to the small area occlusion problems and outperforms the state of art method in several cases.