采用精确抠图的单幅自然图像背景虚化合成方法  

Synthesis Method for Background Blurring of Single Natural Image Using Accurate Matting

作　　者：李晓颖 李思琪 洪雪梅[1] 张海[1] LI Xiaoying;LI Siqi;HONG Xuemei;ZHANG Hai(College of Information Science and Engineering,Huaqiao University,Xiamen 361021,China;Xiamen Acdemy of Arts and Design,Fuzhou University,Xiamen 361021,China)

机构地区：[1]华侨大学信息科学与工程学院,福建厦门361021 [2]福州大学厦门工艺美术学院,福建厦门361021

出　　处：《华侨大学学报（自然科学版）》2023年第3期391-397,共7页Journal of Huaqiao University(Natural Science)

基　　金：福建省自然科学基金面上资助项目(2021J01288)。

摘　　要：针对小光圈镜头下的单幅自然图像背景虚化问题,提出一种基于精确抠图的单幅自然图像背景虚化合成方法.首先,利用GrabCut算法对单幅自然图像的前景对象进行初始抠图;其次,采用基于局部二值模式算子的颜色恒常性算法提取图像纹理,并对初始抠图实现精确割图;然后,通过双边滤波虚化背景部分;最后,将精确分割后的前景部分与虚化后的背景部分合成.仿真结果表明:文中算法可以获得高质量的分割和视觉自然的背景虚化效果.Aiming at the problem of background blurring of single natural image under small aperture lens,a synthesis method of background blurring of single natural image based on accurate matting is proposed.Firstly,the GrabCut algorithm is used to perform initial matting on the foreground object of single natural image.Secondly,the color constancy algorithm based on local binary patterns operator is used to extract the image texture,and the initial matting is precisely cut.Then,the background is virtualized by bilateral filter.Finally,the accurately segmented foreground part is combined with the virtualized background part.Simulation results show that the proposed algorithm can achieve high-quality segmentation and visually natural background blurring effects.

关 键 词：图像虚化 局部二值模式 颜色恒常 GrabCut算法 双边滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]