为解决传统图像修复过程中存在的不一致问题,提出一种基于窄带优化的 自适应多匹配块随机查找图像修复方法。利用小波变换分解破损图像获得 不同分辨率的低频与高频子图。由每一子图破损边界上待修复块颜色和结构 信息自适应计算其匹配块尺寸及数量。基于最小堆随机查找匹配块序列,并 利用窄带模型对其进行优化以修复破损边界。对破损边界由外向内逐层修复直至 破损区域补全,用小波回复算法重构出最终的修复结果。实验结果表明与 已有方法相比该方法的修复结果具有更好的视觉一致性。

In order to solve the inconsistent problem of traditional image inpaint process, an adaptive multiple matched patches random search image completion method based on narrowband optimization is proposed. Damaged images are decomposed by wavelet transformation to obtain low and high frequency sub-images with different resolution. The size and number of matched patches are adaptively calculated according to color and structure information of unrepaired patches along the damaged boundary of each sub-image. Matched patch sequence is found randomly based on the minimum heap, and the narrowband model is used to optimize the damaged boundary. The damaged boundary is inpainted from outside to inside until the damaged area is completed. The final inpainted result is restructured using the wavelet reconstruction algorithm. Experimental results show that the completion result with the proposed method has better visual consistency compared with the existing methods.