字典学习优化结合HMAX模型的鲁棒人脸识别

摘要

Since the variation of attitude and illumination in face recognition may affect on the recognition performance,a face recognition method combining dictionary learning optimization with HMAX model is proposed. In the method,the combina⁃tion of sample image and affine package model got from the sample image is used to express an image;after that the HMAX model is used to extract the C2 features,and the eigenmatrix is optimized with dictionary learning;and then the C2 features of vi⁃sual attention model and original model are combined,and classified with support vector machine(SVM). The results of experi⁃ment relied on Caltech and AR face database show that,in comparison with other advanced face recognition methods,the pro⁃posed method has better recognition performance,and has the robustness for the variation of facial expressions and illumination.%针对人脸识别中由于姿态、光照等变化而影响识别性能的问题，提出了字典学习优化结合HMAX模型的人脸识别方法。首先，使用样本图像和从样本获得的仿射包模型联合表示一幅图像；然后，利用HMAX模型提取C2特征，并利用字典学习优化特征矩阵；最后，将视觉注意模型与原始模型的C2特征进行组合，并利用支持向量机完成分类。在Caltech和AR人脸数据库上的实验结果表明，相比其他几种较新的人脸识别方法，提出的方法取得了更好的识别性能，对人脸表情和光照变化具有鲁棒性。