Sparse Reconstruction in Compressed Sensing with Block-Based Approach

In this paper, we focus on the sparse reconstruction problem in compressed sensing. While Matching Pursuit (MP) based algorithms represent a common approach, we identified limitations in handling large-scale problems. To address this, we propose a novel block-based methodology. Specifically, we partition the atom dictionary into multiple blocks and conduct performance evaluations through systematic testing. This approach significantly reduces computational complexity by processing subsets of atoms iteratively (e.g., implementing block-wise correlation calculations instead of full-dictionary operations) while maintaining reconstruction accuracy. Our experimental results demonstrate that this block-based method exhibits substantial advantages for large-scale problems, making it suitable for practical applications. The implementation typically involves dividing the dictionary matrix into submatrices, applying MP iterations per block, and aggregating results using intelligent merging strategies.