基于破损区域分类的自适应扩散模型

为了提高数字图像的修复精度,加快修复速度,文章提出了一种自适应扩散模型。该模型利用结构张量的最大特征值将破损图像区分为平坦区域、结构信息不明显的区域以及结构信息明显的区域,使模型在不同的区域进行相应的扩散；此外,文章利用结构张量的两个特征值构造了速度函数,使模型在不同的区域的扩散速度也不同,从而加快扩散速度并保持图像的结构信息。实验结果表明,该模型不仅可以加快修复速度,也可以很好地抑制噪声,并且保持图像的结构特征。

Adaptive diffusion model based on damaged region classification

In order to not only improve the accuracy of digital image inpainting, but also to accelerate inpainting rate, this paper proposed an adaptive diffusion model. According to maximum eigenvalue of structure tensor the flat region, this model can distinguish less obvious structure region and the obvious structure region. And in different regions, the model can diffuse in the corresponding ways. Meanwhile, using two eigenvalues of structure tensor, this paper constructed smooth velocity, so that in different regions, the model had different diffusion speed, which could speed up the diffusion and maintain the structural characteristics. Experimental results show that, the model can not only accelerate repainting, but also can denoise effectively, and maintain the structural characteristics of the image.