核Fukunaga-Koontz变换检测红外小目标  

DETECT INFRARED POINT TARGETS BASED ON KERNEL FUKUNAGA-KOONTZ TRANSFORM

作　　者：刘瑞明[1] 刘尔琦[2] 杨杰[1] 张田昊[1] 王芳琳[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200240 [2]中国航天科工集团公司第二研究院,北京100854

出　　处：《红外与毫米波学报》2008年第1期47-50,55,共5页Journal of Infrared and Millimeter Waves

基　　金：国家自然科学基金(60675023,60602012);国防973项目基金(JC2000020100104)

摘　　要：Fukunaga-Koontz变换作为一种两类模式识别工具,源自主成分分析,只能提取图像的二阶统计特征,没有考虑图像的高阶统计特征,用于红外小目标检测时难以达到满意效果.为了提高目标检测性能,本研究首先将Fuku-naga-Koontz变换推广到高维核特征空间,成为核Fukunaga-Koontz变换,以提取图像的高阶统计特征,从而利用核Fukunaga-Koontz变换来检测红外小目标.Fukunaga-Koontz transform （FKT） is a pattern recognition tool for two-class problems. It can only capture the second-order statistical features of images without considering high-order statistical properties because it stems from the principal component analysis （PCA）. So the satisfying detection performance can hardly be achieved. For improving the detection performance, Fukunaga-Koontz transform was firstly extended to the higher-dimensional space in order to capture the higher-order statistics of images in this study. And it is named as kernel Fukunaga-Koontz transform. The kernel Fukunaga- Koontz transform can be used to detect infrared point targets.

关 键 词：Fukunaga-Koontz变换 红外小目标 目标检测 主成分分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]