Adaptive Rate Block Compressive Sensing Based on Algorithmic Reconstruction Error

Compressive sensing (CS) for a full-sized image requires a large sensing matrix, consuming significant storage space and computational resources. Block-based CS addresses this but allocates the same number of samples to each block, ignoring the sparsity differences among various blocks. Additionally, the original image's sparsity is unknown in practical systems. Therefore, it is necessary to accurately estimate the sparsity from some initial sampling information. This paper proposes an adaptive rate block CS model based on the difference of the results of two reconstruction algorithms. The image is initially sampled at the sampling end. Then a high-accuracy algorithm and a low-accuracy algorithm are used to reconstruct initial sampled signals respectively. The sparsity of each image block is estimated by the difference between the two initial reconstructed signals, allowing the appropriate allocation of the number of samples for a block. The main computations of this scheme are concentrated at the reconstruction end, which can effectively save resources at the sampling end. Experimental results show that the proposed scheme can effectively improve reconstruction quality while reducing the sampling rate.