基于局域熵值分布图的目标分割及质心计算  

Target segmentation and centroid calculation based on local entropy distribution

作　　者：赵齐月 毛征[1] 张庆龙[1] 孟凡刚[1] 曲劲松[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124

出　　处：《国外电子测量技术》2014年第2期33-36,48,共5页Foreign Electronic Measurement Technology

摘　　要：图像分割的质量直接决定了目标检测及跟踪等一系列工作的好坏。传统的局域熵对于简单背景下的小目标检测较为有效,但对于目标较大的情况分割效果则不甚理想。致力于统计图像每一像素点的局域熵值,绘制局域熵值分布图。对局域熵值分布图进行迭代阈值分割,以求得最佳分割阈值并实现二值化。在二值化熵值分布图的基础上,统计像素分布,求取目标质心。在下一帧的视频图像中,根据前一帧求取的目标质心来设定波门,以减少非目标以及噪声的干扰。对多个视频序列进行实验后证明,本文提供的方法能够有效减少杂波干扰,提高系统的鲁棒性。The quality of image segmentation directly determines the effect of target detection and tracking.The traditional entropy detection method about small target on simple background is effective,but the segmentation result for a larger target is not very ideal.This paper is devoted to the statistics of the entropy value of each pixel and local entropy distribution map drawing.Using iterative threshold method segments entropy distribution map,in order to obtain the optimal segmentation threshold and realize binaryzation.Based on the binaryzation image,calculate the pixel distribution and the centroid of the object.In the next frame of the video,we can set a searching window according to the object centroid seeked from the previous frame,in order to reduce the non-target and noise interference.Experiments on several video sequences demonstrate that this method can effectively reduce the interference of clutter,improve the robustness of the system.

关 键 词：图像分割 局域熵 迭代阈值法 质心 

分 类 号：TN957.522[电子电信—信号与信息处理]