基于CDD模型的快速图像修复算法

A Rapid Image Inpainting Algorithm Based on CDD Model

为对图像的缺损部分进行快速自动修复,提出了一种基于曲率驱动修复模型的快速图像修复算法。曲率驱动修复模型由于引入了曲率项,使其偏微分方程为高阶,修复时需要数值求解偏微分方程,大量迭代运算导致修复速度非常缓慢。为加快修复速度,算法将模型的偏微分方程数值化,进一步改造成加权平均形式,利用邻近已知像素直接合成损坏像素,加权系数由曲率和梯度共同确定,使修复按照图像等照度线方向进行,在曲率大的地方将等照度线拉伸,同时由待修复点邻域内已知像素的梯度方差确定修复次序。实验结果表明,显著减小了运算时间,一定程度满足"连接性准则",并且对于较小破损区域修复效果好于曲率驱动修复模型。

In order to restore the damaged domain in image automatically and rapidly, a rapid image inpainting algorithm based on Curvature Driven Diffusions（CDD） model is proposed here. Because the Partial Differential Equation （PDE） of CDD model is high order equation as its curvature item,in which,much iterative calculation is in need to numerically solve the PDE and this leads to much time consuming. To inpaint the damaged image rapidly, a weighted mean equation is got by modifying the numerical PDE of CDD model firstly ,then the new pixel is produced by interpolating the damaged pixel with its neighborhood pixels weighted mean, the weighted factors are decided by gradient and curvature items. This can make interpolation processed according to isophote direction, and the isophote lines will be extended on large curvature positions,finally the inpainting order is decided by variance of gradient. The experimental results show that this algorithm is much faster in computation than CDD model and realizes the connectivity principle partially. In addition ,the method has better result in inpainting small damaged domain compared with CDD model.