Abstract: In complex scenarios,illumination change,occlusion and noise make the image intra-variance of recognition algorithm (taking pixel feature as similarity measure) greater than the between-class variance,and reduce the perfor-mance of face recognition.To solve this problem,this paper proposed an low-rank supported extreme learning machine for robust face recognition to improve recognition performance.Firstly,the subspace linear assumption of face image distribution is used to make the image waiting to be recognized cluster to the corresponding sample subspace.Secondly,the pixel domain is resolved into low-rank feature subspace and sparse error subspace,and the forward network of low-rank structure characteristic of face image for training extreme learning machine is extracted,according to the low-rank principal of the image subspace for noise robustness.Finally,the extreme learning machine face recognition algorithm for noise robustness is realized.Experimental results show that,compared with the state-of-the-art face recognition algorithm,the proposed method not only has high recognition accuracy,but also has lower time complexity and better practicability.

Key words: Face recognition,Noise robust feature,Low-rank matrix recovery,Extreme learning machine