复杂场景多传感器图像的多目标分割算法  

Multi-target Segmentation in Multi-sensor Images in Complex Scenes

作　　者：马超杰[1] 杨华[1] 李晓霞[1] 凌永顺[1] 吴丹[1] 王静雯[1] 

机构地区：[1]电子工程学院安徽省红外与低温等离子体重点实验室,合肥230037

出　　处：《光电工程》2009年第1期47-51,共5页Opto-Electronic Engineering

基　　金：安徽省红外与低温等离子体重点实验室基金(2007A011011F)资助项目

摘　　要：综合应用多传感器图像灰度分布特征和分形维数特征进行目标检测、应用动态边缘演化提取目标轮廓,有效解决了复杂场景多目标分割的难题。首先根据图像的内在属性应用指定直方图进行图像增强,随后分别对可见光图像应用分形维数特征、对红外图像应用最大熵法、对激光雷达图像应用局部阈值法进行兴趣区域提取;再将各传感器图像获得的兴趣区域进行交叉验证,进一步排除背景干扰;最后把经过综合处理的目标轮廓估计作为初始生长曲线应用动态边缘演化技术最终确定目标边缘。对大量的复杂场景多传感器图像测试表明,本文提出的方法较好地保留了目标的形状特征,是一种有效的多目标分割技术。The multi-target segmentation in complex background could be solved effectively by synthetically utilizing multi-sensor images＇ gray level distribution, fractal dimension and active contour evolution technology. Firstly, according to image inherent attribution, the images were enhanced using the histogram specification. Then the regions of interest were selected in multi-sensor images by using fractal dimension features for visible images, maximum entropy method for infrared images and the local threshold for Lidar images. The regions-of-interest obtained in multi-sensor images were across verified, so background clutter could be further eliminated. Finally, the target contour estimation could be used as the initial growth curve for active contour evolution processing, and the better target contours were obtained on the true target boundaries. A large number of segmentation tests on multi-sensor images in .complex scenes prove the validity and reliability of the scheme

关 键 词：多传感器 多目标分割 分形特征 动态边缘演化 复杂场景 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]