Lasso正则化Gabor剪切波人脸多元稀疏函数逼近

LRGS:The Multiple Sparse Function Approximation Based Lasso Regularization Gabor Shearlets for Face Recognition

针对传统人脸识别方法中识别精度不高,对人脸数据中存在的方向性和各异性特征处理效果不佳的问题,提出一种基于Lasso正则化的Gabor剪切波人脸多元稀疏函数逼近算法。首先,在人脸图像生物信号层面,利用Gabor改进剪切波算法,对人脸数据特性进行稀疏膨胀表示,并利用其对具有方向性和各向异性特性的人脸膨胀几何特征进行提取;其次,为平衡算法效果,引入Lasso正则化理论来控制和权衡人脸数据的保真度和平滑度间的关系;最后,通过与已有算法在标准测试库上的仿真实验对比,验证了所提算法在人脸识别精度以及效率上的优势。

According to the problem of low accuracy in face recognition with traditional algorithm, here proposed a multiple sparse function approximation based lasso regularization Gabor shearlets for face recognition. Firstly, in the biological signal level, here used the Gabor improvement shearlets to representation the face characteristics data in sparse method, and also used it to extract the face geometric features with its directionality and anisotropy; Secondly, here introduced the regularization theory to control and measure the relationship between the fidelity and smooth of face data; Finally, through the experiments on the standard test with existing algorithms showed the proposed algorithm was more accuracy and efficiency in the recognition results.