摘要
巴氏距离可以衡量离散概率分布的相似性.在Komodakis的基于置信传播的图像修复算法基础上,提出在搜索最佳匹配块时,以欧式距离为主、巴氏距离为辅的方法度量待修复块与样本块之间的相似性,以提高正确匹配率.结果表明,该方法对缺失纹理的修复具有很好的效果,明显提高了修复质量.
Bhattacharyya distance measures the similarity of discrete probability distribution.On the basis of Komodakis' priority belief propagation scheme,a novel algorithm using Euclidean and Bhattacharyya distance to measure the similarity between the missing patch and sample patch was proposed to improve matching correctness in a better way.The experimental results showed that the method had good performance in recovering the missing textures and achieved impressive inpainted results.
出处
《上海应用技术学院学报（自然科学版）》
2014年第3期228-232,共5页
Journal of Shanghai Institute of Technology: Natural Science
基金
国家自然科学基金资助项目(61401281)
上海市自然科学基金资助项目(14ZR1440700)
上海市高校青年教师培育基金资助项目(ZZyyy13022)
上海应用技术学院引进人才基金资助项目(YJ2013-10)
关键词
图像修复
图像补全
相似性度量
欧式距离
巴氏距离
image inpainting
image completion
similarity metrics
Euclidean distance
Bhattacharyya distance
