符合人类视觉感知的图像对象分割方法  

Image Object Segmentation Method in Accord with Human Vision Perception

作　　者：陶超[1] 谭毅华[1] 蔡华杰[1] 田金文[1] 

机构地区：[1]华中科技大学图像识别与人工智能研究所,武汉430074

出　　处：《计算机工程》2010年第24期6-8,共3页Computer Engineering

基　　金：国家"863"计划基金资助项目(2007AA12Z153);民用航天预研计划基金资助项目

摘　　要：提出一种符合人类视觉感知的图像对象分割方法,包括双尺度的区域分割和基于模型的对象提取。运用非线性尺度算子对图像进行大尺度平滑,结合颜色量化和视觉一致性的颜色聚类完成图像的粗分割。在原尺度上融合区域的纹理、颜色信息对分割区域进行区域合并,并利用对象模型完成图像对象的提取。实验结果表明,该算法的分割结果符合人类视觉感知特性,能够较好地完成图像对象分割。Image object segmentation method in accord with human vision perception, which consists of double-scale region segmentation and object extraction based on object model, is proposed in this paper. An image is smoothed by nolinear scale operator in coarse-scale, color quantization and perceptual color clustering are exploited to form the initial segmented regions. The initial regions are merged based on the region distance in the original finest-scale. Object model is used to implement image object extraction. Experimental results demonstrate that the proposed algorithm gets better image object segmentation with favorable consistency to human vision perception.

关 键 词：视觉感知 对象分割 双尺度空间 区域合并准则 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]