组稀疏学习在图像去噪中显示出巨大潜力, 但现有方法仅从图像块级别考虑含噪图像的非局部自相似性, 影响了强噪声图像的重建质量. 本文在组稀疏复原模型中引入组稀疏残差和全变分正则化约束, 将含噪图像复原转化为多尺度图像块匹配和减少组稀疏残差的问题. 之后, 基于干净图像的组稀疏系数预估和多尺度图像块匹配, 提出了有效的自适应图像复原迭代算法, 提升组稀疏学习算法的图像去噪和精细结构复原能力. 实验结果表明, 本文方法在强噪声图像复原的主、客观综合评价上优于BM3D、WNNM等标杆去噪算法.

Compressed sensing based on group sparsity has shown great potential in image denoising. However, most existing methods considered Nonlocal Self-Similarity (NSS) prior of noisy images only in a block-wise manner, which reduced reconstruction quality. This paper introduced group sparsity residual and total variance as the supplemental constraint within the framework of compressed sensing based on group sparsity, and transformed the reconstruction problem into two issues: multiscale patch matching and decreasing group sparsity residual. Then, an effective iterative algorithm with adaptive regularization parameter was proposed to recover the noisy images after estimating original images’ group sparse coefficients and matching patches at multiple scales, which improved group sparsity learning’s performance in denoising and restoring fine structure. Experimental results demonstrated that the proposed algorithm outperforms the contrast benchmarking algorithms for images corrupted with strong noise, such as BM3D, WNNM when considering the visual results and the objective evaluation together.