基于特征融合的图像目标显著性检测方法  

Image Target Saliency Detection Method Based on Feature Fusion

作　　者：李德峰 刘松涛[2] LI Defeng;LIU Songtao(The 27th Institute of China Electron.Technol.Group Corporation,Zhengzhou 450015,CHN;Dept.of Information Operation,Dalian Naval Academy,Dalian 116018,CHN)

机构地区：[1]中国电子科技集团公司第27研究所,郑州450015 [2]海军大连舰艇学院信息作战系,辽宁大连116018

出　　处：《半导体光电》2018年第6期898-902,908,共6页Semiconductor Optoelectronics

基　　金：国家自然科学基金项目(61303192);博士后基金项目(2015M572694;2016T90979)

摘　　要：为了凸显图像中的感兴趣目标,提出了基于特征融合的图像目标显著性检测方法。首先通过提取可见光图像不同尺度空间的不同特征,利用区域协方差理论融合尺度空间之间串接的不同特征,然后结合全局核密度估计体现图像的全局显著性,实现局部和全局特征融合的图像目标显著性检测。仿真结果表明,无论主观评价,还是客观指标,新方法均优于当前流行的图像显著性检测方法。In order to make the interest target prominent,a saliency detection method based on feature fusion is proposed.Firstly,by extracting the different characteristics of different scale space in the visible image,different concatenated characteristics among scale space were fused by using the theory of regional covariance;finally by combining the global kernel density estimation, the global saliency of the image was reflected.Thus the image target saliency detection was realized by fusing the global and local features.Experimental results show that,regardless of the subjective assessment,or the objective indicators,the proposed method is superior to usual saliency detection methods.

关 键 词：图像目标 显著性检测 区域协方差 全局核密度估计 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]