基于流形特征与形状先验的红外直升机图像分割  

Infrared Helicopter Image Segmentation Based on Manifold Feature and Shape Priori

作　　者：周则明 胡彪 孟勇 陈超迁 罗其祥 ZHOU Ze-ming;HU Biao;MENG Yong;CHEN Chao-qian;LUO Qi-xiang(College of Meteorology and Oceanography,National University of Defense Technology,Nanjing,Jiangsu 211101,China)

机构地区：[1]国防科技大学气象海洋学院,江苏南京211101

出　　处：《电子学报》2018年第4期834-839,共6页Acta Electronica Sinica

基　　金：国家自然科学基金面上项目(No.61473310;No.41174164);国家自然科学基金青年科学基金项目(No.41305138)

摘　　要：为有效分割复杂天空背景下的直升机目标,提出了基于流形特征与形状先验的变分分割模型.根据图像数据的灰度分布构造区域能量项,推动变形曲线向目标边界演化;引入对称正定(Symmetric Positive Definite,SPD)矩阵流形上的区域协方差描述子构造流形特征能量项以提高分割算法的鲁棒性.在区域项、边界项和流形特征项的共同作用下获取红外直升机目标的第一阶段分割结果.在第二阶段分割过程中,基于主成分分析(Principal Component Analysis,PCA)方法获取直升机目标的先验形状变化模式,以变形曲线在PCA空间重构的形状作为先验知识约束曲线的演化,最终实现红外直升机图像的分割.实验结果表明,本文方法能够有效获取直升机目标的完整轮廓.In order to segment the helicopter target in the infrared image with complex cloud background effectively,a variational segmentation model based on manifold feature and shape priori is proposed.The region energy term is constructed on the basis of gray distribution of the image data todrive the deformation curve to target boundary.A regional covariance descriptor is introduced on Symmetric Positive Definite(SPD)matrix manifold to define aprior feature energy term to improve the robustness of the segmentation algorithm.The segmentation result of the helicopter in the first stage of the infrared image is obtained by the combination of the region term,the boundary term and the prior feature term.In the second stage,Principal Component Analysis(PCA)is used to capture the prior shape variation pattern of the infrared helicopter target,and the shape of evolution curve is reconstructed in PCA space as the prior knowledge to restrain the curve deforming.Finally,the complete contour of the infrared helicopter is obtained.Experimental results demonstrate that this method can effectively extract the whole contour of the infrared target.

关 键 词：红外图像分割 对称正定矩阵流形 区域协方差描述子 先验形状 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]