Image compressed sensing recovery using intra-block prediction

Considering the strong correlation among the neighboring areas in an image, in this paper, we propose a compressed sensing recovery of images by providing an intra-block prediction for the image. Starting from an initial, direct compressed-sensed reconstruction and splitting it into non-overlapping blocks, a prediction for every block within the image is built up based on the average of its most similar neighboring blocks. An enhanced recovery of the image is obtained by adding the reconstructed residual-obtained as difference between the measurements of the original image and the obtained measurements of its prediction image-to the generated prediction. Experimental results show that the proposed algorithm outperforms, both subjectively and quantitatively, the direct reconstruction using the initial measurements alone.