非线性各向异性扩散的图像修复

Image inpainting based on nonlinear anisotropic diffusions

根据图像曲率和梯度的特性引入扩散控制变量因子,提出非线性各向异性扩散的图像修复方法,能根据图像本身的几何信息进行不同方向和不同强度的扩散。其中关键参数p能根据图像局部几何信息的曲率和梯度自适应地改变,并控制扩散方向和扩散强度。扩散过程中,在图像的边缘区域,沿边缘方向扩散具有较大的扩散系数,沿垂直边缘方向扩散具有很小的扩散系数;在图像的平坦区域,向周围等强度扩散,而且扩散强度值较大。实验结果与经典的全变分、曲率驱动和P-laplace常数变分方法做比较,表明研究方法能对图像的破损区域进行修复,提高图像的质量。

Following the features of curvature and gradient at image edges,an image inpainting method based on nonlinear anisotropic diffusion is proposed. The diffusion can be conducted with different directions and different intensities according to the geometric features of the inpainted images. An adaptive factor is introduced based on the curvature and gradient of the image local geometric information,which can control the diffusion direction and diffusion intensity. At the edges in images,the diffusion coefficients are large for the horizontal directions,while the diffusion coefficients are small for the vertical directions. At the smooth regions,the diffusion coefficients are the same for different directions and they are usually large.Compared with the typical total variation method,the curvature derivation diffusion method,and P-laplace constant variation method,the experimental results show that the proposed method can improve the qualities of the inpainted images