纹理结构引导的自适应图像修复算法

Adaptive Image Inpainting Algorithm Based on Texture Structure

利用约束纹理合成的方法来解决大面积缺损图像的修复问题,可以在视觉上获得比较好的效果。而修复过程中的合成顺序,对最后的结果有很大影响。Criminisi提出的基于样本图像的图像修复算法,在前人基础上,对合成顺序给出了解决方案。文中首先对决定合成顺序的优先级进一步做了改进,在数据项中加入结构张量,使图像修复从结构区域向无结构区域填充。其次,根据原图像区域纹理结构信息,使得每一次合成的块的大小自适应变化,这在一定程度上避免了纹理块过小或过大带来的弊端,从而使合成效果更为自然。经过对比实验,文中算法取得了较好的效果。

Taking into account the large defect area of images, the constrained texture synthesis is used to image inpainting. The filling order is important because it has a great influence on the final result. Based on the previous studies,Criminisi algorithm gives a solution to the filling order. Firstly, an improvement is made on the priority of the filling order in this paper. Structure tensor is added to the data term in order to make the filling order is from the structural region to the non-structural region. Secondly, according to the texture structural information, it makes the size of template block change adaptively. To some extent, it avoids the disadvantages caused by the block of fixed size, so that the result is more natural. The contrast experiment shows that the algorithm obtains better visual appearance.