| 引用本文: | 何仕文,刘琳,张永强,杨剑哲,石大明,程丹松.改进TV-H-1模型的图像修复方法[J].哈尔滨工业大学学报,2016,48(2):167.DOI:10.11918/j.issn.0367-6234.2016.02.029 |
| HE Shiwen,LIU Lin,ZHANG Yongqiang,YANG Jianzhe,SHI Daming,CHENG Dansong.An improved image inpainting method based on TV-H-1 model[J].Journal of Harbin Institute of Technology,2016,48(2):167.DOI:10.11918/j.issn.0367-6234.2016.02.029 |
|
| 摘要: |
| 为改善现存图像修复算法在修复时存在的"灰度跳变"现象，同时降低运行复杂度，提出一种基于偏微分方程模型（称为Isophote-TV-H-1模型）和改进Criminisi算法的数字图像修复算法.首先利用图像分解模型（TV-H-1）获得缺损图像的结构部分和纹理部分；然后用Isophote-TV-H-1模型和改进的Criminisi算法分别对缺损图像的结构部分和纹理部分进行修复；最后将修复后的结构部分和纹理部分进行叠加得到最终的修复结果.实验结果表明，本模型与TV模型相比，能够较好地修复缺损区域中的纹理信息；与Criminisi算法相比，本模型通过对相似度度量方法的改进，有效地抑制了图像修复过程中的误差传播，并利用局部搜索（图像局部相似性）来替代传统的穷尽搜索，进而提高算法的效率.同传统的基于图像分解的图像复原算法以及TV模型相比，本模型能解决"灰度跳变"问题，获得更好的修复结果.
|
| 关键词: 全变分 TV-H-1 图像分解 图像修复 Criminisi算法 |
| DOI：10.11918/j.issn.0367-6234.2016.02.029 |
| 分类号:TP391 |
| 文献标识码:A |
| 基金项目:国家自然科学基金 (5,3); 国家博士后科学基金(20100480998); 国防科工局重大专项(公开)(50-Y20A08-0508-15/16); 哈尔滨市科技创新人才专项资金 (2013RFQXJ110). |
|
| An improved image inpainting method based on TV-H-1 model |
|
HE Shiwen, LIU Lin, ZHANG Yongqiang, YANG Jianzhe, SHI Daming, CHENG Dansong
|
|
(School of Computer Science and Technology, Harbin Institute of Technology, 150001 Harbin, China)
|
| Abstract: |
| Intensity discontinuity and high computational complexity are drawbacks in some existing methods of image inpainting. To tackle these problems, a method based on PDE model(Isophote-TV-H-1 model) and improved Criminisi algorithm is proposed in this paper. Firstly, the damaged image is decomposed into cartoon and texture with the TV-H-1 model. Secondly, the Isophote-TV-H-1 model and the improved Criminisi algorithm are used to recover the cartoon and texture of the damaged image, respectively. Finally the recovered texture is superimposed on the recovered cartoon to get the result image. The experimental results demonstrate that the proposed model recovers the texture of the damaged region better than the TV model. Comparing with Criminisi algorithm, the proposed model suppresses the error propagation through improving the similarity measurement method, as well as improves the efficiency by employing the local search.
|
| Key words: total variation TV-H-1 image decomposition image inpainting criminisi algorithm |
