对光照变化鲁棒的快速关键点提取与匹配  

Fast keypoint extraction and matching robust to illumination changes

作　　者：刘天亮[1,2,3] 戴修斌[4,3] 陈昌红[1,2] 朱秀昌[1,2] 刘峰[1,2] 罗立民[3] 

机构地区：[1]江苏省图像处理与图像通信重点实验室,南京210003 [2]南京邮电大学通信与信息工程学院,南京210003 [3]东南大学影像科学与技术实验室,南京210096 [4]南京邮电大学地理与生物信息学院,南京210003

出　　处：《东南大学学报（自然科学版）》2012年第3期413-418,共6页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金青年基金资助项目(61001152);国家自然科学基金资助项目(61071166;61071091;61172118);江苏省自然科学基金资助项目(BK2010523);江苏省高校自然科学基金资助项目(11KJB510012);南京邮电大学校科研基金资助项目(NY210053;NY210069;NY210073;NY211030)

摘　　要：利用积分图像提出一种对光照变化鲁棒的快速关键点提取与匹配方法.首先,对基于黎曼积分的对比度拉伸响应,利用积分图像进行多尺度上采样滤波,快速提取光照鲁棒的局部特征,并在多分辨率框架下基于局部极大值检测多尺度显著特征关键点.然后,使用积分图像方法对以特征点为中心对称的矩形区域快速构造协方差描述,并采用对数欧氏距离,测度2个匹配场景中所提取的关键特征点周围区域的协方差矩阵之间的差异;最后,根据稀疏匹配策略,完成2个特征点集之间的自动匹配.实验结果表明,该方法结构简单,计算快速高效,准确性较高,且对于光照变化具有较好的适应性.An integral images-based fast keypoint extraction and matching method being robust to illumination changes is presented. First, robust illumination local features are fast extracted by multi- scale up-sampling and filtering the Riemann integral of the stretched contrast responses using integral images, and multi-scale interest feature keypoints are detected based on local maximum of them in a multi-resolution framework. Secondly, covariance descriptors are constructed with integral images for the center-symmetrical rectangular regions around the obtained points, then the Log-Euclidean metrics are adopted to measure the difference between the covariance matrices around the keypoints from two matching scenes. Finally, automatic matching betweentwo feature point sets is implemen- ted with the sparse matching strategy. Experimental results prove that the proposed method is of sim- ple structure, fast and efficient computation with high accuracy, while being adaptive for illumination changes.

关 键 词：关键点提取 特征匹配 光照变化 协方差描述子 积分图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]