摘要
目前抠图算法分为采样法和仿射法。采样法求解出的Alpha值往往是不连续的,并且含有很多噪声。对此提出一种基于KNN后处理的鲁棒抠图算法。对未知像素点进行鲁棒性采样,并从中选择较好的样本对作为未知像素的前景和背景样本点,计算出未知像素的初始Alpha值;充分利用KNN方法搜索样本范围宽的优势对初始Alpha进行后处理。实验表明该后处理算法不仅增强了Alpha的视觉效果,同时也提高了Alpha的结果,尤其是实体前景和背景像素结果的准确性。
At present,image matting algorithm is divided into sampling-based method and affinity-based method.The Alpha values obtained by sampling-based method are often discontinuous and contain a lot of noise.To solve the above issues,this paper proposes a robust image matting algorithm based on KNN post-processing.It performed robust sampling of unknown pixels,and the better sample pairs were selected as the foreground and background sample points of unknown pixels,and the initial alpha value of unknown pixels was calculated.The advantages of KNN method were fully utilized to search the sample range and post process the initial Alpha.Experiments show that the post-processing method not only enhances the visual effects,but also improves the accuracy of Alpha results,especially that of opaque foreground and background pixels.
作者
白杨
姚桂林
Bai Yang;Yao Guilin(School of Computer and Information Engineering,Harbin University of Commerce,Harbin 150028,Heilongjiang,China)
出处
《计算机应用与软件》
北大核心
2020年第9期170-175,共6页
Computer Applications and Software
基金
黑龙江省自然科学基金项目(F2018021)
哈尔滨商业大学校级科研项目(17XN058,17XN059)。
关键词
采样抠图法
仿射抠图法
鲁棒抠图方法
KNN搜索
抠图后处理
Sampling-based matting
Affinity-based matting
Robust matting
KNN searching
Image matting post processing
