基于抽风矢量场的深度凹陷图像分割算法  

Segmentation algorithm of recessed image based on vector field of suction

作　　者：孔勇奇[1] 潘志庚[2] 

机构地区：[1]浙江商业职业技术学院信息技术学院,浙江杭州310053 [2]杭州师范大学数字媒体与人机交互研究中心,浙江杭州310036

出　　处：《浙江大学学报（工学版）》2014年第6期1024-1033,共10页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61170318)

摘　　要：为了解决深度凹陷图像分割中的分割瓶颈问题,提出一种基于抽风矢量场的深度凹陷图像分割算法.该算法对凹陷图像进行初次图像分割,获取初次分割曲线及图像梯度矩阵;依据图像梯度矢量的分布规律,对分割曲线的控制节点进行分割瓶颈检测,并以分割瓶颈的弦线中点为中心坐标,引入算法预置的抽风矢量场;以图像梯度标识矩阵为参考,对抽风矢量场中的梯度矢量进行点积运算,并将矢量的点积结果作为分割曲线的外部图像驱动力,实现分割曲线在凹陷区域内部的继续收敛.实验数据显示;在不同凹陷程度情况下,该算法始终能够将图像分割的平均误差和覆盖比率控制在有效范围.不同算法的分割对比实验表明,该算法分割曲线对深度凹陷结构的拟合程度优于其他同类算法.An image segmentation algorithm based on the vector field of suction was presented to solve the bottleneck problem for segmenting recessed images. This method segments the recessed image and saves the image original segmentation curves and the image gradient matrix. It identifies and marks the bottlenecks on the control nodes of the segmentation curve, while the identification process is based on the distribution of gradient vector. Meanwhile, a predefined vector field of suction is introduced and the midpoint of the straight segment on the arc is selected as the coordinates of the center chord on the vector field. By referring the mark matrix on the gradient, the algorithm performs dot product on the gradient vectors in the vector field of suction, and the dot product of the vectors is specified as the drive force for splitting the curve. Thus this method can achieve segmentation curve convergence inside the boundary concavities. Experimental results show that, for varying degrees of boundary concavities. The proposed algorithm can always limit the image segmentation average error and coverage ratio within a valid range. Results of comparative experiments show that the curve obtained by the proposed algorithm is closest to the edge of the image among those obtained by the similar segmentation algorithms.

关 键 词：分割瓶颈 抽风矢量场 梯度标识矩阵 凹陷图像 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]