Image compressive sensing reconstruction based on collaboration reduced rank preprocessing

Abstract

The problem of image compressive sensing (CS) preprocessing is considered. Currently, image CS reconstruction algorithms mainly consider the sparsity prior knowledge of original image. However, the change of the sparsity strength among the different images may degrade the efficiency of the reconstruction algorithms. Thus the idea of CS preprocessing is proposed to serve two purposes: strengthen the sparsity property of the CS measured image and make preprocessing and reconstruction algorithm matched. Specifically, the collaboration reduced rank (CRR) preprocessing is proposed based on non-local sparsity and non-local low-rank regularisation reconstruction algorithm (NLR-CS). Then a more efficient CRR-NLR-CS CS reconstruction method is proposed which utilises the CRR preprocessing and NLR-CS. Experimental results show the effectiveness of the proposed method.