Residual domain dictionary learning for compressed sensing video recovery

Abstract

For compressed sensing (CS) recovery, the reconstruction quality is highly dependent on the sparsity level of the representation for the signal. Motivated by the observation that the temporal residual image is much sparser than its original image, a temporal residual-domain dictionary learning method for CS video recovery is proposed in this paper. The adaptive basis is learned from inter-frame differences by Karhunen---Loeve transform (KLT) to represent the residuals. And a block-based motion estimation/motion compensation (ME/MC) residual reconstruction strategy is incorporated for the CS video recovery. Experimental results on common test sequences at various sampling rates illustrate that the proposed algorithm gains great improvements over existing approaches. For some video sequences, the proposed method outperforms the state-of-art method near 1 dB in terms of peak signal noise rate (PSNR) at some higher sampling rate.