基于多正则化约束的图像去运动模糊

发布时间：2018-03-08 14:11

本文选题：去运动模糊　切入点：多正则化约束　出处：《工程科学与技术》2017年03期 　论文类型：期刊论文

【摘要】：针对图像去运动模糊问题的病态性,已有的方法通常引入对图像的正则化约束从而缩小解空间范围使其良态化,但单一的正则化约束并不能很好地估计点扩散函数和复原原始图像。基于此,本文提出一种基于多正则化约束的图像去运动模糊方法。首先,根据图像梯度符合重尾分布的特性,采用归一化的超拉普拉斯先验项作为对图像先验约束的正则项。其次,分析描述图像运动模糊的点扩散函数的内在特性包括稀疏性和连续光滑性;同时,采用点扩散函数自身的L1范数保证其稀疏性并作为其中一项点扩散函数先验约束的正则项,采用Tikhonov正则化约束保证其连续平滑性并作为另一项点扩散函数先验约束的正则项,避免估计的点扩散函数中存在孤立的点。由于所建立的正则项虽然不可微但其是非严格凸函数,故引入辅助变量采用分裂法和交替求解法对所建能量方程进行求解,并利用小波软阈值公式求解辅助变量。本文方法对合成的运动模糊图像和实际相机抖动造成的自然模糊图像均进行实验,实验结果验证了该模型和求解算法的有效性和快速性。实验结果表明,本文方法提高了点扩散函数估计准确度,同时提高了复原图像质量,具有较好的复原效果。
[Abstract]:In view of the ill-condition of image demotion blur problem, the existing methods usually introduce regularization constraints on images to narrow the solution space and make them better. However, a single regularization constraint can not estimate the point diffusion function and restore the original image very well. Based on this, this paper proposes an image de-motion blur method based on multiple regularization constraints. According to the characteristics of image gradient matching heavy-tailed distribution, the normalized super-Laplace priori term is used as the regular term for image priori constraint. The inherent properties of point diffusion function describing image motion blur include sparsity and continuous smoothness, meanwhile, the L _ 1 norm of point diffusion function is used to guarantee its sparsity and to be a regular term which is a priori constraint of point diffusion function. The Tikhonov regularization constraint is used to guarantee its continuous smoothness and to be a priori constraint of another point diffusion function. In order to avoid the existence of isolated points in the estimated point diffusion function, the established canonical terms are not differentiable but they are not strictly convex functions, so the auxiliary variables are introduced to solve the energy equation by splitting method and alternating solution method. Using the wavelet soft threshold formula to solve the auxiliary variables, both the synthetic motion blur image and the natural blur image caused by the camera jitter are tested in this paper. The experimental results verify the validity and rapidity of the model and the algorithm. The experimental results show that the proposed method improves the accuracy of point diffusion function estimation and the quality of restored images.
【作者单位】：成都信息工程大学计算机学院;成都信息工程大学图形图像与空间信息协同创新中心;
【基金】：国家重点基础研究发展计划资助项目(2014CB360506) 四川省科技支撑计划资助项目(2015RZ0008) 四川省教育厅重点项目资助(15ZA0186) 国家自科基金青年基金资助项目(61303126)
【分类号】：TP391.41

【相似文献】