Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (8): 189-194.DOI: 10.3778/j.issn.1002-8331.1712-0311

Previous Articles     Next Articles

Restoring Method of Defects Based on Continuous Patch of Structure and Multi-Scale Transform

SHI Xiaobo, LIN Suzhen, SANG Xinya   

  1. School of Computer and Control Engineering, North University of China, Taiyuan 030051, China
  • Online:2019-04-15 Published:2019-04-15



  1. 中北大学 计算机与控制工程学院,太原 030051

Abstract: Aiming at the problem of poor continuity of information after the repairing of serious defects, the ancient tombs mural images are taken as an example, a restoration method based on the combination of continuous blocks and multi-scale transformation is proposed. Firstly, the defect boundary is determined according to the result of image defect marking. Secondly, the priority of boundary point is determined by decreasing confidence, matching confidence, data item and gradient term. Thirdly, it uses the pixel and structure difference between the block to be repaired and the matching block global search to get the best matching block to replace the to-be-repaired block. Then, the new-fill-point confidence item is updated, the defect boundary is re-determined until empty, the coarse repair result is obtained and converted into HSV color space. Finally, the Non-Subsampled Contourlet Transform(NSCT) is used to decompose the [V] component, the low frequency information is inversely transformed with the high frequency information after being restored by the suppression data item method, the result is combined with the [H] and [S] components and converted into the RGB color space to obtain the final image restoration result. The experimental results show, compared with Criminisi and other methods, the method based on the combination of continuous blocks and multi-scale transform has enhanced continuity, peak signal-to-noise ratio and mean-square error, proving the effectiveness of the proposed method.

Key words: image inpainting, tomb murals, Non-Subsampled Contourlet Transform(NSCT), continuous patch of structure

摘要: 针对严重缺损图像修复后往往存在信息连续性差的问题,以古墓葬壁画图像为例,提出基于结构连续块和多尺度变换相结合的修复方法。依据图像缺损标注结果确定缺损边界;通过下降置信度项、匹配置信度项、数据项和梯度项确定边界点优先权;利用待修复块与匹配块间像素差异和结构差异全局搜索得到最佳匹配块以替换待修复块;更新新补点置信度项,重确定缺损边界直至为空,得到粗修复结果,并转换到HSV(Hue Saturation Value)色彩空间;采用非下采样轮廓波变换(Nonsubsampled Contourlet Transform,NSCT)分解[V]分量,低频信息用抑制数据项方法修复后与其高频信息逆变换,其结果与[H、][S]分量合成并转换到RGB色彩空间,得到图像最终修复结果。实验结果表明,相比Criminisi等方法,基于结构连续块和多尺度变换相结合的方法在结构连续性、峰值信噪比和均方根误差方面均有所提升,证明了所提出方法的有效性。

关键词: 图像修复, 古墓葬壁画, 非下采样轮廓波变换, 结构连续块