一种抗视角变换的改进SIFT景象匹配算法  

Improved SIFT algorithm on scene matching for anti-viewpoint changing

作　　者：苏可心[1,2] 韩广良[2] 孙海江[2] 

机构地区：[1]中国科学院大学,北京100039 [2]中国科学院长春光学精密机械与物理研究所图像室,长春130033

出　　处：《计算机工程与应用》2015年第5期172-177,共6页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61172111)

摘　　要：针对光电成像制导过程中,电视导引头实时获取的前视图像与预存的目标区域基准图像之间存在较大视角差异时,SIFT(Scale Invariant Feature Transform)算法得到的正确匹配点对数较少导致匹配失败的情况,提出一种改进的SIFT景象匹配算法。该算法通过在高斯尺度空间中提取抗仿射变换区域,并进行区域归一化处理来获得抗视角变换特征点集的方法,使正确匹配点对明显增加。实验结果表明,该算法对视角变换具有很好的鲁棒性:在视角差高达50°至60°以上时得到的匹配准确率较SIFT算法提高近5倍,具备工程实用价值。To overcome the great viewpoint changing between the real-time image which is acquired by the seeker and the reference image of target area during the electro-optical imaging guidance, an improved SIFT algorithm on scene matching is proposed to bridge the gap of the original one, which may lead to failed scene matching for the lack of correct matched pairs. It increases matched pairs by extracting viewpoint changing resistant points which got from the affine transformation resistant regions extracted in the Gaussian scale-space. Experimental results show that the proposed algorithm is robust to viewpoint changing：it can acquire nearly 5 times more correct rate of matching than the original one with the viewpoint ranges from 50° to more than 60° and can satisfy the practical need.

关 键 词：尺度不变特征变换(SIFT) 抗仿射变换区域 视角变换 景象匹配 

分 类 号：TP752[自动化与计算机技术—检测技术与自动化装置] TN911.73[自动化与计算机技术—控制科学与工程]