Research On Face Recognition With Low Quality Images Based On Sparse Representation

Face recognition has important research value and application significance in pattern recognition.However,in the real world,face images are often affected by illumination,disguise,corruption and occlusion,namely,low quality face images.Traditional face recognition methods are difficult to perform well for face recognition with low quality images.Thus,considering feature extraction and classifier algorithm,this paper proposes Sparse Low-rank Component Representation(SLCR)and Sparse Low-rank Component Coding(SLC)for face recognition with low quality images,which are based on Sparse Representation-based Classification(SRC)and can reduce the impact of interference information in the training set.The main work of this paper includes:(1)Propose a Sparse Low-rank Component Representation(SLCR):(a)Utilize low-rank component and non-low-rank component of the training dataset to describe the effective feature and the other information associated with occlusion,outlier and sparse error.Since face images within a class have a low-rank structure,low-rank component can describe the effective feature better.(b)Propose a Sparse Low-rank Component Representation(SLCR).SLCR utilizes low-rank matrix recovery on the training dataset to obtain low-rank component and non-low-rank component,which construct the dictionary.Then,SLCR uses the Augmented Lagrange Multiplier(ALM)scheme to obtain the solution.Finally,SLCR minimizes class-wise reconstruction residual to recognize the testing image.(2)Propose a Sparse Low-rank Component Coding(SLC): Based on the theory of robust M estimation,a new residual function is designed to reduce the impact of the outliers for face recognition with illumination and corruption.Based on sparse low-rank component representation,this paper uses new residual function to overcome the influence of illumination and noise on the training dataset for face recognition with low quality images.This paper's methods are compared with the other SRC-based methods on the Extended YaleB,CMU Multi-PIE,AR,CAS-PEAL and LFW face databases' different low quality images dataset.The results show that this paper's methods not only are suitable for face recognition,but also outperform the other SRC-based methods for face recognition with low quality images.