On prediction error compressive sensing image reconstruction for face recognition

This paper explores the scope of spatial domain sparse representation for the application to develop a fast and robust remote end face recognition (FR) scheme in the framework of compressive sensing (CS). At the source end, error images as the difference between the original and the predicted images, are obtained using the different predictors that offer compressive measurements. Sub-sample measurements of the sparse error image and part of the original image are then transmitted. At the destination end, the test image is obtained from its partial information and CS reconstructed error image. Principal Component Analysis is used to extract the important features from the reconstructed image followed by FR. Performance of the proposed method is studied using collaborative representation based classifier with regularized least square method, applied on two databases, AR and ORL and an accuracy of 93.99% for the former and 91.5% for the latter is observed.