面向小目标图像的快速核密度估计图像阈值分割算法  

Fast Kernel Density Estimator Based Image Thresholding Algorithm for Small Target Images

作　　者：王骏[1] 王士同[1] 邓赵红[1] 应文豪[1] 

机构地区：[1]江南大学数字媒体学院,无锡214122

出　　处：《自动化学报》2012年第10期1679-1689,共11页Acta Automatica Sinica

基　　金：国家自然科学基金(60903100);江苏省自然科学基金(BK2009067;BK2011417);中央高校基本科研业务费专项资金(JUSRP111A38);江苏省信息融合软件工程技术研究开发中心开放基金项目(SR-2012-01)资助~~

摘　　要：针对当前小目标图像阈值分割研究工作面临的难题,提出了快速核密度估计图像阈值分割新方法.首先给出了基于加权核密度估计器的概率计算模型,通过引入二阶Renyi熵作为阈值选取准则,提出了基于核密度估计的图像阈值分割算法(Kernel density estimator based image thresholding algorithm,KDET),然后通过引入快速压缩集密度估计(Fastreduced set density estimator,FRSDE)技术,得到核密度估计的稀疏权系数表示形式,提出快速核密度估计图像阈值分割算法fastKDET,并从理论上对相关性质进行了深入探讨.实验表明,本文算法对小目标图像阈值分割问题具有更广泛的适应性,并且对参数变化不敏感.In order to threshold the image containing small targets well, a novel fast kernel density estimator based image thresholding algorithm is proposed. Firstly, a novel computation model for probability density estimation based on the kernel density estimator with weighting coefficients is presented. By introducing the 2nd-order Renyi entropy as the threshold selection criterion, a novel kernel density estimator based image thresholding algorithm （KDET） is proposed. Then a fast version for KDET, named fastKDET, is proposed by integrating fast reduced set density estimator （FRSDE） and RSDE into the data condensation procedure. Moreover, some fundamental theoretical properties are fully studied. At last, several experiments are conducted and show that our fastKDET is more general than some of the existing algorithms and is insensitive to the parameters.

关 键 词：核密度估计 快速压缩集密度估计 阈值分割 小目标图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]