摘要
数字抠像是将一幅图像中的前景物体与背景进行分离的问题,它的关键在于Alpha通道的计算.以往通过采样方法求得的Alpha中,由于采用逐点计算的离散化方式,求解出的Alpha通常不连续,并且包含很多噪声,因而需要对Alpha进行后处理,这不仅会增强Alpha在视觉上的平滑性,而且能够进一步提高Alpha的精确度.在目前国际上,有关数字抠像后处理领域已经进行了许多研究,但缺少相关的综述性文献,并且对后处理后的Alpha如何进行定量的评价也仍未系统解决.本文首先将数字抠像中的后处理方法分为2类:与仿射类方法相结合的方式及自平滑方式,其次,对两类方法进行了全面的总结和梳理,并对方法的优缺点进行了分析,对将来研究方向提出了建议,最后,针对后处理后的Alpha结果进行了全面的量化比较,弥补了传统方法基本上仅在视觉层面上进行比较的缺陷.
Image matting is a process which separates the foreground object from the background scene, and the key of matting is to compute the alpha matte. The existing sampling based matting methods are always in a discretized strategy, which could results in a great deal of discontinuities and noises in final alpha mattes. Post processing is thus introduced to en- hance the smoothness and to further increase the accuracy of the final matte. However, the corresponding review articles are still lacking in the field of international research of post-processing in image matting. Moreover ,the quantitative evaluation of alpha mattes still remains unsolved. This paper firstly classifies the post-processing step into two basic categories: affinity- combined and self-smoothing. Next, the advantages and disadvantages are both summarized and analyzed. Finally, the alpha mattes after post-processing are evaluated in quantitative manner comprehensively, which improves the problem of pure visual evaluation in traditional methods.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2017年第3期719-729,共11页
Acta Electronica Sinica
基金
国家重点基础研究发展计划项目(No.2015CB351804)
国家自然科学基金(No.61472103)
国家自然科学基金重点项目(No.61133003)
哈尔滨商业大学博士科研启动项目(No.15KJ06)