摘要
提出了一种红外序列图像中小目标的Dempster-Shafer(DS)证据理论识别方法。DS证据理论在图像融合中存在两个实际问题:一是经典的DS证据理论组合算子在处理高冲突信息时会产生违反常理的结果,二是如何将图像特征转化为证据理论中各个命题的基本概率指派。针对这两个问题,引入证据距离,改进证据组合公式,解决了高冲突证据下的信息融合问题,保证了融合算法的可靠性;运用模糊逻辑,求出了图像中各像素隶属于目标的基本概率指派。对实际的红外序列图像运用改进的公式进行像素级融合,并根据最大概率准则确定出图像中的小目标。实验结果表明,新方法更为有效、可靠,识别错误率可降低到6.14%。
A method of image fusion with Dempster-Shafer （DS） evidence theory is proposed to recognize small targets in infrared image sequence. There are two practical problems of DS evidence theory in image fusion： one is that classical evidence theory involves counter-intuitive behaviors when the highly conflicting information exists, the other is how to determine the basic mass function of the hypothesis in evidence theory. To solve these problems, a robust method based on modified evidence theory is proposed and the mass functions can be obtained from the characters of infrared images. The proposed method is applied to small target recognition. Experimental results show that the new method is more effective and reliable with error-rate 6.14%.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2005年第8期6-8,31,共4页
Opto-Electronic Engineering
基金
国家自然科学基金(0400067)
上海市自然科学基金(03ZR14065)
国防重点实验室基金
关键词
目标识别
红外目标
图像融合
图像分割
Target recognition
Infrared target
Image fusion
Image segmentation
