基于多准则采样的自然图像抠图算法  

Natural Image Matting Based on Multi-Criterion Sampling

作　　者：苟洪山 杨圆 张远 吴磊 谭棉 GOU Hong-shan;YANG Yuan;ZHANG Yuan;WU Lei;TAN Mian(School of Mathematical Sciences and Information Engineering,Guizhou Minzu University;School Guizhou Key Laboratory of Pattern Recognition and Intelligent System,Guizhou Minzu University,Guiyang 550025,China)

机构地区：[1]贵州民族大学数据科学与信息工程学院 [2]贵州民族大学贵州模式识别与智能系统重点实验室,贵州贵阳550025

出　　处：《软件导刊》2023年第2期28-34,共7页Software Guide

基　　金：贵州省科技计划项目(黔科合基础ZK[2022]195);贵州省青年科技人才成长项目(黔教合KY字[2021]104)。

摘　　要：抠图作为图像处理领域的热点难题,已广泛应用于目标识别、虚拟现实和前景提取等方面。然而,图像高质量像素较少时难以获取高质量像素对的问题一直存在,鉴于此,提出基于多准则采样的抠图算法。该算法通过多准则采样策略从全局相似性到局部相似性的角度获取高质量的候选子集,从而解决高质量像素对较少时易丢失等问题。在此基础上,设计多评价联合的最优像素对选取策略,该策略通过联合多个像素对评价函数,以避免单一评价函数估计的解不是当前最优解。为了验证该算法优越性,选用Alpha matting基准数据集作为实验数据。实验结果表明,在图像高质量像素较少时难以获取高质量像素对的问题上,该算法估计前景透明度遮罩优于目前典型的抠图算法。As a hot research problem in the field of image processing, image matting is widely used in target recognition, virtual reality and foreground extraction. However, it is always difficult to obtain high-quality pixel pairs when there are few high-quality pixels in the image.Therefore, this paper proposes an image matting algorithm based multi criteria sampling, it uses a multi criteria sampling strategy to obtain high-quality candidate subsets from the perspective of global similarity to local similarity, to solve the problem that high-quality pixel pairs are easy to lose when there are fewer high-quality pixel pairs. On this basis, a strategy for selecting the optimal pixel pair with multiple evaluations is designed,it combines multiple pixel pair evaluation functions to avoid the problem that the solution estimated by a single evaluation function is not the current optimal solution. In order to verify the superiority of the image matting algorithm based multi criteria sampling, the Alpha matting benchmark dataset is selected as the experimental data. The experimental results show that, compared with the current popular matting algorithm, the foreground alpha mattes estimated by the proposed algorithm is optimal.

关 键 词：多准则采样 联合评价函数 采样抠图算法 图像处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]