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-Curvature Driven D iffusions( CDD) model is proposed he re. Because the Partial D iffe rentia l Equa􀀁

tion( PDE) of CDD model is high order equation as its curvature item, in wh ich, much iterative ca lculation is in need

to num er ica lly so lve the PDE and th is leads tom uch tim e consum ing. To inpa int the dam aged im ag e rapidly, a we igh􀀁

ted m ean equa tion is got by mod ify ing the num erica lPDE of CDD model firstly, then the new p ixe l is produced by in􀀁

terpo la ting the dam aged pix elw ith its ne ighborhood pixels w eigh ted m ean, the we ighted factors are decided by grad ient

and curvature item s. This can m ake interpo la tion pro cessed accord ing to isophote d irection, and the isopCurvature Driven D iffusions( CDD) model is proposed he re. Because the Partial D iffe rentia l Equa􀀁

tion( PDE) of CDD model is high order equation as its curvature item, in wh ich, much iterative ca lculation is in need

to num er ica lly so lv