空间目标图像的非凸稀疏正则化波后复原

Non-convex sparsity regularization for wave back restoration of space object images

现有的空间目标图像波后处理方法多直接套用自然光学图像的复原技术,效果并不理想。本文通过分析空间目标图像的近似稀疏性和灰度值服从超拉普拉斯分布的独有特点,提出了一个采用正则化方法的非凸稀疏正则化空间目标图像复原模型。在数值计算过程中,根据交替方向乘数法将复原模型分解为两个子问题,对凸优化子问题采用快速傅里叶变换求解,对非凸优化子问题采用固定点迭代方法求解。文中设计了非凸稀疏正则化空间目标图像波后复原的完整算法流程,并针对模拟图像和真实空间目标图像进行了对比验证。结果显示:相对于最近的流行算法,提出方法的最大峰值信噪比提高了2dB,最大平均结构相似度提高了0.17,最大信息熵提高了3.85,图像清晰度提高了2.65。

The wave back restoration of space object images is usually performed by restoration methods for nature optical images,however,the restoration effect is not ideal.This article proposes a restoration model of a space object image based on non-convex sparsity regularization according to the approximate sparsity of the space object image and the features that the gray value submits to Hyper-Laplace distribution in a regularization way.With the alternating direction multiplier method,the restoration model is split into two sub-problems in the numerical solving process：Fast Fourier transformation is used to solve the convex sub-problem,while the fixed-point iteration is used to solve the nonconvex sub-problem.Then,it gives a complete process for the proposed wave back restoration method of space object images,and do an experiment to test and verify the simulated images and the real spaceobject images.Compared results show that proposed method improves the largest peak signal to noise ratio by 2dB,the average structural similarity by 0.17 and the information entropy and the image definition by 3.85 and 2.65,respectively.