基于快速视网膜局部特征的遥感图像目标识别  

Remote sensing image target recognition based on fast retina key point local invariant feature

作　　者：陈彦彤[1,2] 徐伟[1] 朴永杰[1] 陈娟[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,长春130033 [2]中国科学院大学,北京100049

出　　处：《仪器仪表学报》2016年第4期852-859,共8页Chinese Journal of Scientific Instrument

基　　金：国家863计划(2012AA121502)项目资助

摘　　要：针对目前航天遥感图像信息复杂、数据量大,导致目标识别中特征检测准确度低、特征匹配识别时间长的问题,提出了一种基于差分加速分割角点检测算法(AGAST-Difference)和快速视网膜关键点描述算法(FREAK)相结合的目标识别方法。在特征检测阶段,建立AGAST-Difference特征检测算子,将尺度空间理论融合到加速分割角点检测算法(AGAST)中,生成具有强仿射不变性的特征点;再利用简化的FREAK采样模型描述局部特征,并构建二进制特征向量,通过计算向量间的汉明距离,完成特征匹配及目标的快速识别;最后选用美国Quick Bird卫星的遥感图片进行验证,实验结果表明,所提特征检测算子仿射不变性能较强,不仅提高了检测的可重复率,而且特征描述符区分性较强,平均匹配正确率提高了9.91%,识别用时仅为35 ms。该方法识别效率高、速度快,能够满足遥感图像实时识别的需求。Aiming at the problems that aerospace remote sensing image contains complex information and large amount of data, which leads to low feature detection accuracy and long feature matching recognition time in target recognition, a new target recognition method is proposed based on AGAST-Difference algorithm and FREAK description algorithm. Firstly, in the feature detection phase the AGAST- Difference feature detector is created, which fuses scale space theory into adaptive and generic comer detection based on the accelerated segment test （AGAST） detection algorithm by adding difference of Gaussian （DoG） operator, and generates the feature points with strong affine invariant. Then, the brief FREAK sampling model is used to describe the local features and build binary feature vectors. Finally, the feature matching and fast target recognition are achieved by computing the hamming distances among the features. The Quick Bird satellite remote sensing images were adopted to verify the proposed method; and the experiment results show that the proposed feature detector has strong affine invariant; it not only improves the detection repeatability, but also has strong discrimination ability for the feature descriptor. The average matching accuracy increases by 9.91%, and the recognition time is only 35ms. To sum up, the proposed method is fast and efficient, and could satisfy the requirement of remote sensing image real time recognition.

关 键 词：图像处理 遥感目标识别 AGAST—Difference检测子 FREAK描述符 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH166[自动化与计算机技术—计算机科学与技术]