A novel hybrid compressed sensing image reconstrcution method

In this paper, a hybrid approach of image reconstruction from highly incomplete data is introduced. The method is a weighted recursive filtering procedure. At each iteration, random noise is first injected in the unknown portion of the spectrum, and then a reweighted denoising filter consisting of Block Matching 3D (BM3D) filter and multiscale L₀-continuation filter is exploited to attenuate the noise in the image domain and reveal new features and details, finally those new features are projected onto the unknown portion of the spectrum to update the K-space data. The proposed method avoids local solutions and recovers the features and details of the image efficiently by utilizing advantages of both filters. The experimental results on both simulated and real images consistently demonstrate that the proposed approach can efficiently reconstruct the image with high image quality.