Spatially adaptive image reconstruction via compressive sensing

Compressive sensing (CS) is an emerging model-based framework for signal recovery at a rate significantly below the Nyquist sampling rate. The CS theory states that a signal having a sparse representation in some bases can be reconstructed from a small set of random projections. In this paper, a reconstruction method is developed based on block CS and adaptive choice of frame expansions according to spatial features of partitioned regions. Natural image is divided into different types of regions, the discrete cosine transform (DCT) is used for smooth regions, while the bi-orthogonal wavelet transform (OWT) is chosen for uneven regions. Several experiments are conducted on benchmark images to verify the efficacy of the proposed method. Experimental results show that it achieves improved quality in both subjective and objective measurement as compared with existing methods.