摘要
已有的协同分割方法没有考虑到同一类图像所具有的目标形状相似性,从而使得分割结果不一致。提出了形状模板约束的图像交互协同分割算法,通过少量用户交互提高协同分割质量。该算法首先定义形状模板;然后通过形状上下文实现分割结果传递,自动形成图像分割所需的前景和背景掩码;最后采用最小割理论进行分割边界优化。实验结果表明,与已有的协同分割算法相比,该算法能在简单用户交互下明显提高分割质量,使分割结果更具有语义性。
Existing image co-segmentation method fails to consider the similarity of shape between images,so that the segmentation results are inconsistent.We introduced a method of interactive image co-segmentation by the constraints of shape.First,the method preprocess an input image and finds the template of shape.Secondly,shape context matching algorithm is used to produce foreground and background mask for graph cut.Finally,minimum cut theory is used to optimize segmentation boundary.The experimental results show that compared with existing image co-segmentation method,this algorithm can significantly improve the quality of segmentation and make the segmentation results more semantical.
出处
《计算机科学》
CSCD
北大核心
2016年第11期309-312,共4页
Computer Science
基金
省自然科学基金项目(LY15F020024
LY16F020033)
浙江省文物局(2014014)资助
关键词
图像协同分割
形状模板
形状上下文
图割
Image co-segmentation
Shape template
Shape context
Graph cut
