非凸高阶全变差正则化自然光学图像盲复原

受噪声和图像边缘结构信息的影响,传统的图像盲复原方法易出现"振铃"、"拖尾"、"阶梯"等现象。为解决上述问题,本文利用图像的后验信息、点扩散函数(PSF)的稀疏性以及l1,l2两类范数在约束中的不同作用,提出了一种更一般的非凸高阶全变差正则化自然光学图像盲复原模型。针对提出模型的非凸优化问题,在数值求解过程中对模型的范数结构进行改进,引入Split-Bregman权值迭代方法,提高了计算精度。对人工模拟退化图像和真实图像进行了实验测试。结果表明,提出的方法能够对多种退化类型的图像进行有效复原,复原后的图像边缘保持良好,细节和纹理的处理都优于最近文献提出的模型。客观评价结果显示,相比最近文献的模型,提出模型的峰值信噪比最大可以提高2.08dB,信息熵值最大可以提高1.14个单位。

Blind restoration of nature optical images based on non-convex high order total variation regularization

Influence by noise and image edge structure information, traditional blind image restoration methods usually result in special phenomena of ringing, tail and ladder. To solve these problems, this paper proposes a more general blind restoration model of nature optical images based on non-convex high order total variation regularization by using the posteriori information of an image, the sparse property of a Point Spread Function(PSF) and different advantages of norm l₁ and norm l₂ in restriction. In the numerical solving process, the Split-Bregman iteration method was introduced by improving the norm of the model structure to improve the calculation accuracy and to solve the non-convex optimization. The experimental test between artificial simulation degradation images and real images was performed. Results show that the proposed method restores effectively variety types of degenerated images, and the restored images have well edges and their texture details are better than that of the models in recent literatures. The objective appraisal indicates that the peak signal-to-noise ratio has increased by 2.08 dB and the largest improvement of the information entropy reaches to 1.14 units as compared to the latest literature models.