基于边缘自适应小波变换的图像修复算法

Image Inpainting Based on Edge Adaptive Wavelet Transform

本文提出一种基于边缘自适应小波变换的多尺度图像修复算法,对非纹理图像有比较好的修复效果。边缘自适应小波变换的基本思想是,先检测出图像的主要边缘,这些边缘把图像分割成几个平滑区,然后对图像进行不跨越边缘的小波分解,即在各平滑区内部进行小波变换,得到图像的多尺度表示,并且同时计算边缘的多尺度表示。这样的小波分解使高频信息基本都集中在边缘上,而高频系数则非常稀疏,而且都接近于零。在此基础上进行图像修复,就只需要对低频部分与边缘图像进行修复,然后重构得到修复图像即可。经过小波分解,低频部分破损区域大大缩小,用比较简单的插值方法就可进行修复,大大降低了计算量。对边缘图则可用曲线拟合的方法进行修复。

A fast inpainting algorithm based on edge-adaptive wavelet transform are introduced in this paper.The main idea of edge-adaptive wavelet transform is to firstly detect the main edges of the image and then carry out a wavelet transform not cross those edges.It makes the high-frequency coefficients more sparse and more weak.After this kind of wavelet transform,inpainting only need to fill in the low-frequency sub-band and connect the broken edges.Because the inpainting field become more small in the low-frequency sub-band,it only need a simple interpolation algorithm and the whole inpainting process getting more faster.