一种动态剪枝的协同稀疏表示方法研究及应用

The Dynamic Pruning Collaborative Sparse Classification and Its Application

该文提出一种动态剪枝的协同稀疏表示方法,通过建立2种不同的训练样本筛选策略,再融合2种策略的优点及结合TPTSR框架进行图像识别,以求获得更好的分类效果.在带噪声的人脸数据集上进行对比实验,结果表明:该方法可以在人脸受到遮挡和光照变化的影响下达到更高的识别率,并具有较强的鲁棒性.

The dynamic pruning collaborative sparse classification(DPCSC)method that is two flexible strategies of selecting suitable and competitive training,samples for sparse representation are built separately has been proposed,and also can be combined with TPTSR framework in image recognition for achieving better performance.Extensive experiments conducted on publicly available face datasets with noise clearly show that the proposed DPCSC performs excellent accuracy under the influences of occlusion and illumination variations on the face image,and robustness for face images with illumination and occlusion.