基于带权核范数最小化和混合高斯模型的去噪模型

非局部自相似性（NSS）先验在图像恢复中发挥重要作用，如何充分利用这一先验提高图像恢复性能仍值得深入研究，提出一种基于带权核范数最小化和混合高斯模型的去噪模型。首先，采用混合高斯模型（GMM）对无噪声的自然图像非局部自相似图像块进行训练，再用训练好的混合高斯模型指导退化的图像产生非局部自相似图像块组；然后，结合带权的核范数最小化技术实现图像的去噪，并对模型的保真项进行一般性扩展，给出收敛的求解算法。仿真实验表明，所提方法与基于3D滤波的块匹配（BM3D）算法、同时稀疏编码学习（LSSC）算法和带权的核范数最小化（WNNM）模型相比，峰值信噪比（PSNR）提高0.11~0.49 dB。

Image denoising via weighted nuclear norm minimization and Gaussian mixed model

Nonlocal Self-Similarity (NSS) prioritization plays an important role in image restoration, but it is worthy of further research that how to make full use of this prior to improve the performance of image restoration. An image denoising via weighted nuclear norm minimization and Gaussian Mixed Model (GMM) was proposed. Firstly, the clean NSS image blocks of the natural image were trained by GMM, and then the trained GMM was used to guide the degraded image to produce NSS image blocks. Then, the weighted nuclear norm minimization was used to realize image denoising, an extended model was proposed by modifying the fidelity item, and the corresponding convergent algorithm was given. The simulation results show, compared with some advanced algorithms such as Block Matching with 3D filtering (BM3D), Learned Simultaneous Sparse Coding (LSSC) and Weighted Nuclear Norm Minimization (WNNM), the proposed method improves the Peak Signal-to-Noise Ratio (PSNR) by 0.11 to 0.49 dB.