结合SVM和分形维数的多特征红外人造目标提取  

Extraction of infrared man-made target based on multi-features by combining SVM with fractal dimension

作　　者：张长江[1] 陈源[1] 赵翠芳[1] 杨丽丽[1] 

机构地区：[1]浙江师范大学数理与信息工程学院,浙江金华321004

出　　处：《浙江师范大学学报（自然科学版）》2013年第2期133-139,共7页Journal of Zhejiang Normal University:Natural Sciences

基　　金：浙江省科技厅公益性应用研究计划项目(2012C23027;2012C31005);计算机软件与理论浙江省重中之重学科开放基金资助项目(ZC323011014)

摘　　要：研究了红外图像中人造目标的提取.首先,通过计算红外图像目标的分形维数确定红外目标和背景的大致区域;然后,分别提取目标图像和背景图像的灰度级特征(邻域中心像素亮度、邻域中值亮度和邻域平均亮度),再利用支持向量机(SVM)进行训练,并尝试用不同的核函数及其参数建立最适当的区分目标和背景像素点的模型,进而把红外图像像素点分成目标和背景2类;最后,利用构建的模型实现红外图像中人造目标的提取.实验结果表明,用该方法建立的分类模型可以有效地提取红外图像中的人造目标.It was studied the extraction of man-made target in infrared image,it was first determined by computing the image fractal dimension of the approximate area of the infrared target,and the background were extracted from the grayscale characteristics of the target image and the background image（the pixel brightness of the center of the neighborhood,neighborhood values brightness and neighborhood average luminance）.The support vector machine（SVM） for training,a different kernel functions and function parameters were used to establish the most appropriate model to distinguish between target and background pixels,and then divided into two of the target and background pixelsclass.The built model was the most final extraction of man-made target in infrared image.Experiment results showed that the classification model established by this method could effectively extract the man-made target in infrared image.

关 键 词：SVM 分形维数 红外图像 人造目标 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]