一种基于Gestalt原理的多连通分量形状外包络提取方法  

A Gestalt-Based Method of Envelope Extraction for Multi-Object Shapes

作　　者：吕肖庆[1] 符松平[1] 宋建国[1] 高良才[1] 汤帜[1,2] 

机构地区：[1]北京大学计算机科学技术研究所,北京100080 [2]数字出版技术国家重点实验室(北大方正集团有限公司),北京100871

出　　处：《计算机学报》2013年第10期2070-2077,共8页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(2012CB724108);北京市自然科学基金资助项目(4132033)资助~~

摘　　要：图形识别领域的学者们正在将研究重点从简单图形转向复杂图形,其中,由多个独立图元(该文中称为连通分量)组合成的外轮廓对于复杂图形的整体分析起着至关重要的作用.该文提出了一种针对多连通分量形状提取整体外轮廓(即外包络)的方法.该方法基于德劳内三角剖分(Delaunay Triangulation),根据Gestalt原理,通过对轮廓片段的凹度和连通分量之间松散度的分析,构造了紧致的外包络,特别是针对断口情况,即由于明显内凹或间隙过大造成的视觉缺失,提出了一种基于平滑曲线对接的恢复方法,最终得到了更符合人类视觉感知的外轮廓.实验对象选自MPEG-7图形库,实验结果表明了该文方法在提取复杂图形外轮廓方面的有效性.In recent years, multi object shape analysis, compared to simple-object shape, has received great attention from more and more researchers. The envelope of multi-object shape contains shape information of each of the inner objects （components） in an image, which plays an important role in understanding the image. In this paper we propose an envelope extraction method for multi-object shapes based on Delaunay Triangulation. By analyzing the concavity of contour segments and the porosity among the individual components, a perceptual contour is constructed. To reduce the gap between digital contour and visual perception, especially in the serious concave and large gap zone, we propose an algorithm of smooth connection with Bezier curves to rebuild the gap zone. Several experiments were conducted on MPEG-7 image collections, and experimental results show the effectiveness of our algorithm on multi-object envelope extraction.

关 键 词：外包络提取 多图元形状 平滑对接 Gestalt原理 图形检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]