基于离散小波变换的图像修补方法*

根据多尺度分析原理,提出了基于离散小波变换的图像修复方法.首先用离散小波变换把图像分解为高频成分与低频成分,对图像不同频率的成分分别进行修补.图像的低频部分采用笔者以前所提出的结合中值滤波和基于曲率扩散方法1]进行修复.由于高频部分地表示图像的边缘轮廓信息,并且有很强的方向性,对高频图像数据先进行分块,用线性拟合求出每一块的方向,再根据方向信息进行修复.实验表明,该算法能较好地修补破损区域.

Image inpainting method based on discrete wavelet transformation

According to the multi-resolution analysis rule,the proposed method dealt with image inpainting via the discrete wavelet transformation(DWT).The damaged image was decomposed using the DWT at first,and then inpainted sub-frequen- cy bands separately.The method of blending curvature-driven and median filter used in our previous works was adopted to in- paint the low frequency information.Since the high frequency contains the direction information of the edges in the image,the image was first divided into many small blocks,the line fitting was used to approximate the direction of the edges to each block,and then the values of the damaged pixels in the block were approximated according to this direction.The results of the experiments show that the proposed method works well in the inpainting of the damaged image.