通过非参数核密度估计优化局部图像模糊

Optimize Local Image Blur By Non-Parametric Kernel Estimation

解决这个问题的例子(估计和优化局部图像的模糊)来自一个场景中的移动物体。不同的相机抖动下,对象的边界会突然变得模糊。为了应对这一点,提出一个自动恢复图像的方法,同时确定模糊区域,并估计其模糊内核。在广泛的解决方法中,该模型没有限制候选模糊离散集,但允许任意的非参数模糊核。与以前的方法相比,所提出的局部模糊估计可以通过用一个像素的潜变量来表示有效模糊内核。具体而言,整合潜在的图像衍生工具,允许边缘密度估计的两个模糊内核和区域的影响。在应用对象的运动模糊和散焦模糊去除后获得清晰的图像。一系列专业方法通过两个数据集实践的定性结果表明非参数方法的多功能性和有效性。

Estimation and optimization of local image blur, the problem comes from a moving object in the scene. Under different camera shaking, the border of object will suddenly become blurred. In order to deal with this problem, proposes a method to recover the image automatically for determining the blur region and estimating the blur kernel. In the common range of solutions, our model does not limit candidate blur discrete sets, but allows any non-parametric blur kernel. Compared with the previous, the local blur estimation can be achieved by using a latent variable to represent the effective blur kernel. Specifically, integrates potential image derivative tools that allow the edge density estimation of the two blur kernel and region effects. Obtains clear images after motion blur and defocuses blur removal of the applied object. The qualitative results of a series of professional methods through two datasets demonstrate the versatility and effectiveness of the nonparametric approach.