摘要
直线段是计算机视觉中常用的一种中层符号表示。将对应于同一实际目标结构的那些直线段组织在一起 ,不但可减少符号数量 ,还可利用符号之间的内在关系来纠正一部分错误。根据人类视觉感知组织的规律 ,研究了利用直线段之间的邻近、共线、交会、平行及对称等非偶然特性实现编组的方法。该方法利用多特征融合的手段处理多个证据 ,并用信任函数表示各编组特征对连接、对称或交会等命题的支持程度 ,没有设置任何固定门限。对仿真图像与室外真实图像的实验显示了该算法的良好效果。
Straight line is a kind of intermediate symbol representation usually used in computer vision. By grouping line segments belonging to the same actual structure of an object, one can reduce the number of symbol, and correct some mistakes utilizing the inherence relations among symbols. The paper studied an approach of grouping based on the rules of perceptual organization, making use of un-accidental features such as proximity, collinearity, junction, parallelism, and symmetry. Multi-feature fusion is adopted when more than one evidence exists. A believe function is used to represent the degree to which each feature support the propositions such as connectivity, symmetry and junction. No fixed threshold is set here. Experiments on both synthetic and real images demonstrate the validity and efficiency of this algorithm.
出处
《计算机应用研究》
CSCD
北大核心
2002年第6期64-67,共4页
Application Research of Computers
基金
国家高技术研究发展计划项目 (86 3 30 8)
卫星应用技术重点资助项目 (Y96 12
Y99 17)
