计算机应用 ›› 2012, Vol. 32 ›› Issue (02): 493-503.DOI: 10.3724/SP.J.1087.2012.00493

• 图形图像技术 • 上一篇    下一篇

基于双树复小波域的马尔可夫随机场样本修补算法

王爽1,陈广秋1,宋亚姬1,2,孙俊喜1   

  1. 1. 长春理工大学 电子信息工程学院,长春 130022
    2. 空军航空大学 基础部,长春 130022
  • 收稿日期:2011-07-11 修回日期:2011-09-23 发布日期:2012-02-23 出版日期:2012-02-01
  • 通讯作者: 孙俊喜
  • 作者简介:王爽(1985-),女,吉林长春人,硕士研究生,主要研究方向:模式识别、图像处理;
    陈广秋(1977-),男,吉林长春人,讲师,硕士,主要研究方向:模式识别、图像处理;
    宋亚姬(1984-),女,吉林长春人,助教,主要研究方向:模式识别、图像处理;
    孙俊喜(1971-),男,吉林长春人,教授,博士,主要研究方向:模式识别、图像处理。
  • 基金资助:
    国家自然科学基金资助项目(60772153)

MRF exemplar inpainting algorithm based on dual-tree complex wavelet domain

WANG Shuang1,CHEN Guang-qiu1,SONG Ya-ya2,3,SUN Jun-xi1   

  1. 1. School of Electronics and Information Engineering, Changchun University of Science and Technology, Changchun Jilin 130022, China
    2. School of Electronics and Information Engineering, Changchun University of Science and Technology, Changchun Jilin 130022, China
    3. Department of Basic Course, Airforce Aviation University, Changchun Jilin 130022, China
  • Received:2011-07-11 Revised:2011-09-23 Online:2012-02-23 Published:2012-02-01
  • Contact: SUN Jun-xi

摘要: 为了消除大目标图像修补过程中,修补区域由于累积误差引起的马赛克和振铃效应,提出基于双树复小波域的马尔可夫随机场(MRF)样本修补算法。首先应用双树复小波变换(DTCWT)将待修补图像变换到复频域,通过合理的置信度和数据项计算待修补块的修补顺序;然后应用MRF样本修补算法在不同尺度、不同方向下修补未知区域;最后利用双树复小波逆变换重构图像。实验结果表明,与传统离散小波修补方法相比,双树复小波域MRF样本修补算法能更好地保持修补区域纹理和结构信息。

关键词: 马尔可夫随机场, 样本修补, 离散小波变换, 双树复小波变换, 图像重构, 纹理信息, 结构信息

Abstract: To eliminate the mosaic and "bell" effects due to cumulative errors during large object image inpainting, the Markov Random Fields (MRF) exemplar inpainting based on dual-tree complex wavelet domain was proposed. The image was converted to complex-frequency domain by Dual-Tree Complex Wavelet Transform (DTCWT) and the exemplar inpainting order was computed by rational confidence and data item, the unknown region was inpainted based on multiscale and multiband. The inpainted images were reconstructed by dual-tree complex wavelet inverse transform. The experimental results show that compared with classical discrete wavelet methods, the mosaic and "bell" effects can be avoided and the more favorable textural and structural information can be preserved.

Key words: Markov Random Fields (MRF), exemplar inpainting, Discrete Wavelet Transform (DWT), Dual-Tree Complex Wavelet Transform (DTCWT), image reconstruction, textural information, structural information

中图分类号: