基于仿射方法的图像抠图算法综述  

A Review on Affinity Based Algorithms in Image Matting

作　　者：姚桂林[1] 姚鸿勋[2] 

机构地区：[1]哈尔滨商业大学计算机与信息工程学院,哈尔滨150028 [2]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《计算机辅助设计与图形学学报》2016年第4期678-693,共16页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61472103);国家自然科学基金重点项目(61133003);黑龙江省自然科学基金面上项目(F201424);哈尔滨商业大学科博士科研启动项目(15KJ06)

摘　　要：数字图像抠图是数字图像处理与编辑领域中的一类经典方法,而仿射方法则是图像抠图中与采样法相并列的一个重要分支,较采样类方法的主要特点和优势是计算速度快、对结果的平滑性能好.本文以闭合形式方法中的早期方法为起点,着重阐述了由它们扩展得到的2种重要闭合类方法各自的优点和缺点,包括经典的Matting Laplacian类方法,以及目前非常流行的Nonlocal类方法.实验结果表明,Nonlocal类在前景与背景重合度不高、边界并不复杂的时候效果较Matting Laplacian类方法要好.最后,本文对闭合形式所衍生出的路径类方法及附加类方法也给出了一些未来的发展建议和规划.Digital image matting is a classical method in digital image processing and editing. Affinity based methods, which share the same level with the famous sampling based one, currently play a significant role in image matting domain. The major advantages of affinity based method over sampling are processing speed and smoothing effect of alpha results. In this paper, in consideration of two early attempts in closed form solutions of affinity based methods, we briefly review two newly popular categories of image matting in closed form manner, and analyze their advantages and disadvantages, including the classical Matting Laplacian based method and the recently popular Nonlocal methods. Experimental results indicate that Nonlocal methods are superior to Matting Laplacian based method in condition of much more dissimilarity between foreground and background colors, and the simpler structure of foreground edges. Besides, we also give some sugges-tions and plans to those path-based and additional methods derived from the closed form solutions.

关 键 词：计算机视觉 图像处理 图像分割 图像抠图 仿射类抠图方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]