【摘要】 针对稀疏表示人脸识别算法对姿态变化敏感的问题,提出一种姿态鲁棒的分块稀疏表示人脸识别算法,通过对人脸进行分块表示并利用仿射变换模型对姿态变化建模,提高稀疏表示人脸识别算法对姿态变化的鲁棒性.同时,通过最小化图像分块重构误差来估计仿射变换参数初值,有效提高仿射变换参数估计精度,进而提升人脸识别算法的性能.实验结果表明,本文算法可在一定程度上克服姿态变化造成的对齐误差,比现有相关算法具有更好的姿态鲁棒性和识别性能.更多还原

【Abstract】 Considering that sparse representation based classification(SRC) is sensitive to pose variation,a novel pose-robust face recognition algorithm via part-based sparse representation(PSRC) was proposed.In PSRC,an image was represented by a set of local regions first,and then the affine transformation model was used to model the pose variation.In this way,PSRC can efficiently improve the robustness of pose variation.Meanwhile,to improve the estimation accuracy of transformation parameters,PSRC got an initial estimation value by minimizing the reconstruction error of local patches.Experimental results show that PSRC can compensate for the alignment errors caused by pose variations and thus outperform most state-of-the-art methods.更多还原