基于二维对称Tsallis交叉熵的小目标图像阈值分割  

Thresholding for small target image based on 2-D symmetric Tsallis cross entropy

作　　者：吴一全[1,2,3] 沈毅[1] 刚铁[2] 吉玚[1] 于素芬 

机构地区：[1]南京航空航天大学电子信息工程学院,南京210016 [2]哈尔滨工业大学先进焊接与连接国家重点实验室,哈尔滨150001 [3]光电控制技术重点实验室,洛阳471009

出　　处：《仪器仪表学报》2011年第10期2161-2167,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60872065);先进焊接与连接国家重点实验室开放课题研究基金(AWPT-M04);光电控制技术重点实验室和航空科学基金(20105152026)联合资助项目

摘　　要：现有的阈值分割方法应用于目标与背景面积相差悬殊的小目标图像时,几乎都失效。为此,提出了基于对称Tsallis交叉熵及背景与目标面积差的小目标图像阈值分割方法。对称Tsallis交叉熵准则能确保准确分割时目标和背景内部的灰度均匀,而背景与目标面积差可抑制均等分割的趋势,二者综合构成了更为合理的阈值选取准则函数。首先导出了一维阈值选取公式;然后经推广得到基于二维斜分对称Tsallis交叉熵及背景与目标面积差的阈值选取公式,给出了其快速递推算法及相应的简化方法。大量实验结果表明:与目前性能较优越的二维斜分Otsu、最大熵、非对称交叉熵阈值分割方法相比,所提出的方法在小目标图像分割效果上具有极为明显的优势。When applied to segment the small target image with significant area difference between background and object,existing methods almost fail.Thus,a thresholding method for small target images is proposed on the basis of the symmetric Tsallis cross entropy and the area difference between background and object.The criterion of symmetric Tsallis cross entropy can make the gray level uniform within the object and background for accurate segmentation;and the area difference between background and object is used to inhibit the tendency of an equal division.Therefore,a more reasonable threshold selection criterion is formed comprehensively.Firstly,1-D threshold selection formula is derived;then it is extended to obtain the threshold selection formula based on 2-D histogram oblique segmentation symmetric Tsallis cross entropy and the area difference between background and object.A fast recursive algorithm and its corresponding simplified version are given.Experiment results show that compared with the thresholding methods of the Otsu,the maximum entropy and the asymmetric Tsallis cross entropy based on 2-D histogram oblique segmentation,the suggested method is extremely effective in segmenting small target images.

关 键 词：图像阈值分割 红外小目标 焊缝缺陷 对称Tsallis交叉熵 背景与目标面积差 二维直方图斜分 快速递推算法 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]