近邻样本协作表示的人脸识别算法

Face recognition using collaborative representation with neighbors

在Gabor特征空间,根据相关系数寻找测试图像的近邻样本,并用这些近邻样本构造完备的冗余字典,从而提出一种基于Gabor特征的近邻样本协作表示的人脸识别算法.在l2范数约束下,利用可变厚度的紧致字典对测试图像进行稀疏表示,根据稀疏系数逐类计算重构图像和测试图像之间的误差,并判断测试图像所属类别.该算法在FERET、ORL和AR数据上进行了无遮挡测试,在AR库上进行了有遮挡测试.实验结果表明,无论有无遮挡,识别速度和识别率都得到了明显改善.

An improved face recognition algorithm using the collaborative representation with nearer neighbors of the testing image is proposed. As a measurement to find the neighboring testing sample, the correlation coefficient between the testing sample and training samples is calculated in the Gabor-feature space. Neighbors of the testing sample compose the compact over-completed dictionary which is variable for different testing samples. The testing image is represented collaboratively by the variable ＂ thickness＂ compact dictionary and the sparse representation coefficient is calculated with 12 minimization. The error between the reconstructed image and the testing image categorizes the testing image. This proposed algorithm has been carried out in database of FERET, ORL and AR with variations of lighting, expression, pose, and occlusion. Extensive experiments demonstrate that the proposed approach is superior both in recognition rate and in speed.