摘要
通过对Knockout算法的研究分析,针对其在实际应用中需要用户大量的交互工作以及因模型简单造成的抠图不精确问题,提出了一种改进算法。该算法用区域生长方法将用户输入的前、背景线段生长聚合成前、背景区域;然后用轮廓查找方法找到两个区域的轮廓作为Trimap的前、背景轮廓线;最后使用优化颜色样本的方法对每个未知像素点的样本进行优化,提高抠图的精确度。实验结果表明,改进后的算法需要用户较少的工作量,能够得到传统Knockout算法更好的抠图效果。
With study and analysis on Knockout matting algorithm, aiming at the problem that much users＇ interworking is needed and the inaccurate caused by the simple model, an improved algorithm is proposed. Region growing method is used to grow foreground lines and background lines into two regions, foreground region and background region;then contours finding method is used to produce foreground and background contour of the Trimap ;at last, optimized color sampling procedure is used to optinfize the sample of unknown region pixels in order to improve the matting results. The experiments demonstrate that the improved method needs less users＇ work, and better behavior in the natural image matting.
出处
《世界科技研究与发展》
CSCD
2010年第6期769-771,781,共4页
World Sci-Tech R&D
基金
重庆市科技攻关计划(CSTC
2009AC2143)
