Single Sample Face Recognition Based On Sparse Representation And Low Rank Representation

Sparse representation has shown an attractive performance in the field of machine learning in recent years,the essence of sparse representation is to measure the sparsity of vector.Low rank representation is driven by sparse representation,its essence is to measure the sparsity of matrix.Face recognition based on sparse representation or low-rank representation is an important research topic,the main idea is to predict the identity of the query image according to the minimum reconstruction error.Those methods effectively improve the recognition accuracy of face recognition with enough training images for each object,however,the performance deteriorate much for Single sampler face recognition(SSFR).While SSFR is the basic condition in many practical face recognition applications.From the perspective of mining the sparsity of data,this paper will propose solutions to improve the effectiveness of sparse representation and low-rank representation for SSFR:(1)SSFR based on sparse representation.The performance of sparse representation will deteriorate much for SSFR.To address this issue,a set of position images for each training image is created with.Then,a robust sparse representation(RSR)for SSFR is proposed based on a series of experiments.(2)SSFR based on low rank representation.Low rank representation can not be applied directly for SSFR due to the lack of training image.To address this,a discriminative projective low rank representation(DPLR)for SSFR is proposed with the virtual sample set which is built under the generic learning framework.Finally,the proposed algorithms are evaluated on AR and Extended Yale B face databases.The experimental results demonstrate the effectiveness and robustness of RSR for occlusion and lighting variations of face image,RSR can effectively improve the accuracy of SSFR.The experimental results also show that DPLR can effectively improve the low rank representation available for SSFR and archive better recognition accuracy and time efficiency.