Pigment Clustering Inpainting on Residual Mural by Improved K-means
发布时间：2024-08-09
点击次数：
- 所属单位：
- 信息与控制工程学院
- 发表刊物：
- Journal of Residuals science & Technology
- 关键字：
- 中文关键字：颜料聚类；图像修复；壁画，英文关键字：pigment clustering;inpainting;mural
- 摘要：
- Ancient murals have survived in the past hundredsyears, there areparts of them left with many disease. These murals carry the history information and need to be restored. Image inpainting as an effective restoration technology is usually used to rebuild the damaged information. Ming temple mural is colorful for plenty of pigments, so it is need to find exemplars from different pigment regions. These murals use Chinese traditional painting principle named five elementsinclude earth, water, metal, fire, and wood. So we design a novel mural inpainting system using improved k-means to cluster the mural pigments and optimize the inpainting algorithm to find the proper exemplars. Firstly we choose random k to get the preliminary results, it just take the Euclidean distance without image color feature. So we add Bhattacharyya coefficients to reject the similar clustering; secondly we transfer the image from RGB to Lab space which have more color expressions for Ming temple mural's pigments and the colors have more relationships with each other; thirdly we inpaint the mural's missing portions in Lab space and change the similarity between exemplars by Bhattacharyya distance to get a better result. The improved k-means clustering system save the running time, and the Lab space inpainting with novel distance calculation insure the filling effects. The experimental results show that the improved technique restores the Ming temple mural better and reduce the time consuming.
- 备注：
- 吴萌
- 合写作者：
- 王展
- 第一作者：
- 王慧琴,吴萌
- 论文类型：
- 期刊论文
- 卷号：
- 卷：13
- 期号：
- 期：9
- 页面范围：
- 页：24.1-24.8
- 是否译文：
- 否
- 发表时间：
- 2017-12-01