基于ROMP算法的人脸识别

Face Recognition with ROMP Algorithm

针对用传统方法进行人脸识别的识别率不够高的问题,本文在人脸识别中采用正则化正交匹配追踪算法(ROMP),并把其与基于NN,匹配追踪(MP),正交匹配追踪(OMP)的人脸识别算法进行了对比.该算法能一次从冗余字典中选取多个原子,并能够通过正则化准则对选取的原子进行再次筛选,获得最优的原子.实验结果表明,在不同特征提取方法和训练样本数改变的情况下,基于ROMP的人脸识别算法的识别率优于其他算法.

The rate of face recognition with traditional method is not high enough, so the work uses regularization orthogonal matching pursuit algorithm （ROMP） for face recognition, and compares it with nearest neighbor （NN）, matching pursuit （MP） and orthogoual matching Pursuit （OMP） algorithm. The advantage of the ROMP is that it can select multiple atoms from a redundant dictionary at each iteration and filter it again from selected atoms to get optimal atoms by regularization criterion. Experimental results show that the recognition rate based on ROMP is better than other algorithms in the case of different feature extraction methods and different number of training samples.