Dimension adaptive hybrid recovery with collaborative group sparse representation based compressive sensing for colour images

In this paper, a fast and efficient hybrid method of image compressive sensing (termed as HRCoGSR) is designed which can adaptively acquire grey or colour image and can faithfully recover it speedily. The proposed method combines and utilises the approaches of recovery via collaborative sparsity (RCoS) and group sparse representation (GSR). For fast convergence, Gaussian Pyramid (GP) is constructed at the front-end and then block compressive sensing (BCS) based RCoS recovery is applied. In the second phase, restricted GSR process is carried out for further enhancing the perceptual quality. The collaborative sparsity-based CS solution is an iterative method and intends to improve signal-to-noise ratio (SNR) performance of the recovered image. It simultaneously enforces local 2D and 3D non-local sparsity in adaptive hybrid transform domain. Parametric performance of the proposed HRCoGSR method is tested over variety of standard grey and colour images and compared with seven existing state-of-the-art methods. Experimental results show that the proposed HRCoGSR method is highly efficient and much faster than existing methods. The average computational time taken by the proposed method is only 26% of that of the standard RCoS method and 46% of the GSR method.