凹凸范数比值正则化的快速图像盲去模糊

摘要

Blurry image can be represented as the convolution of a latent image and a blur kernel,so it is an ill-posed problem to solve the kernel and the latent image inversely from a single blurry image.The most effective way to solving ill-posed problem is using cost function with priori term.For blind image deblurring problem,we propose a ratio of convex norm to concave norm as a regularization priori term,which has more sparse representation ability.When solving the model by variable splitting method,we propose L1 norm fidelity term to update high-frequency information of the latent image.At the stage of updating the blurring kernel,we propose a linear increasing weight parameter to estimate the blurring kernel gradually by multi-scale approach from coarse to fine.After obtaining the blur kernel,we use a closed threshold formula to estimate the latent image.This method can obtain high-quality image efficiently.The experimental results demonstrate the ef-fectiveness of the model and the rapidity of the algorithm.%模糊图像可表示为清晰图像和模糊核函数的卷积，由模糊图像恢复出清晰图像，需要同时估计模糊核和清晰图像，因此是一个病态问题。优化含有先验项的代价函数是求解病态问题最有效方法之一。针对图像盲去模糊问题，本研究提出具有更强稀疏表达能力的凹凸范数比值正则化先验项，在用变量分裂法求解模型时，提出用 L1范数保真项更新估计图像，在更新模糊核时，提出使用线性递增权重参数对模糊核按多尺度方法由粗到细逐步估计，当获得模糊核后，利用封闭阈值公式估计清晰图像。该方法能快速得到高质量的清晰图像，实验结果验证了模型的有效性和算法的快速性。