针对传统稀疏表征分类器只有在训练样本足够多时才会对特征变化不敏感的缺点, 提出一种 Shearlet多方向自适应加权融合的稀疏表征人脸识别算法。为了提取局部方向信息并降低特征维数, 首先利用 Shearlet变换对图像进行多尺度、多方向分解, 获得子带系数矩阵, 然后根据子带系数矩阵方差的大小对同一尺度的方向子图按主方向排序, 利用子带系数矩阵的能量和均值特征对排序后的人脸子图进行加权融合, 最后为了使得表征系数矢量具有更为显著的稀疏性, 进一步利用融合特征构造字典。在 ORL、FERET和 YALE人脸库中做了多组实验, 结果表明, 该方法能增强对外界环境变化的鲁棒性, 同时可以提高人脸的识别率。

For the reason of the traditional sparse representation classifier is not sensitive to the changes of characteristics only has enough training samples, a face recognition method based on Shearlet multi-orientation adaptive weighted fusion and sparse representation is proposed. In order to extract the multi-orientation information and reduce the dimension of the features, images are decomposed in multi-scale and multi-direction by using Shearlet, and the subband coefficient matrices are obtained. Then, the directional sub charts on the same scale are sorted in the main direction according to the sizes of the variances of the subband coefficient matrices. Furthermore, using the energy and the mean values of subband coefficient matrices, the face sub charts are weighted fused. Finally, Shearlet multi-orientation feature fusion is applied to construct sparse representation classifiers for sparse representation of coefficient vectors. The experiments taken under the ORL, FERET, and YALE database are used to verify the effectiveness of the proposed method, and the results show that the proposed method can effectively enhance the robustness of the external environment changes and improve the recognition rate.