关键词: 压缩感知, 组稀疏表示, 块分类, 自适应初始阈值, 迭代阈值递减

Abstract: Aiming at the problems that structural similarity based inter-frame group sparse representation (SSIM-In-terF-GSR) algorithm can＇t fully utilize the high-quality reconstructed key frame information when reconstructing the smooth region and the sparse processing threshold setting is unreasonable，block classification-based adaptive threshold adjustment group sparse reconstruction (BC-ATA-GSR) algorithm was proposed in this paper． Firstly，image blocks were classified into smooth blocks and motion blocks according to the motion state of the objects in the blocks，and reasonable reference frames were allocated for different types of blocks to improve the reconstruction quality of the smooth regions in the video sequence． Then，in order to retain the sufficient structural information，the initial threshold of sparseness was set adaptively according to the sampling rate and the image block type． Fina-lly，an iterative threshold gradient reduction scheme was proposed to accelerate the iterative convergence rate and improve the quality of reconstruction． Compared with SSIM-InterF-GSＲ algorithm，BC-ATA-GSR algorithm achieves better reconstruction effect，and the average PSNR of the reconstructed QCIF and CIF video sequence are increased up to 3. 77dB and 2. 28dB respectively，and the time complexity is reduced up to 42. 08%．

Key words: compressed sensing, group sparse representation, block classification, adaptive initial threshold, iterative threshold decreases