融合几何特征的压缩感知SIFT描述子  

Compressed sense SIFT descriptor mixed with geometrical feature

作　　者：赵爱罡[1] 王宏力[1] 杨小冈[1] 陆敬辉[1] 何星[1] 

机构地区：[1]第二炮兵工程大学目标探测与图像制导实验室,陕西西安710025

出　　处：《红外与激光工程》2015年第3期1085-1091,共7页Infrared and Laser Engineering

基　　金：国家自然科学基金(61203189);总装预研基金(9140A01060411JB4701)

摘　　要：为了解决尺度不变特征变换(SIFT)描述子在存在较多相似结构的匹配中,易造成误匹配,并且维数较高、匹配耗时的问题,提出了一种融合相对几何位置的压缩感知描述子。首先,以特征点为中心,将周围关键点的相对几何位置(RGL)信息形成尺度和旋转不变的RGL描述子,其次,对SIFT描述子利用压缩感知(CS)理论进行降维,形成CS-SIFT描述子,最后将两者融合形成RGL-CS-SIFT描述子。实验结果表明:与SIFT和PCA-SIFT描述子相比,匹配速度有所提升,匹配准确率明显提高。In order to solve the problems that SIFT（Scale Invariant Feature Transform, SIFT） descriptor may result in a lot mismatches when an image has many similar structures and its high dimensions will consume much time in image matching. This paper presents a compressive sensed SIFT descriptor which is mixed with relative geometry location. At first,this method centers on feature point,and transforms the information of relative geometry location related to around key points into a RGL（Relative Geometrical Location, RGL） descriptor, which is invariant to scale and rotation. Secondly, CS-SIFT（Compressive Sense SIFT, CS-SIFT）descriptor is formed by reducing dimensions of SIFT descriptor using the theory of compressive sense. At last, two descriptors form a RGL-CS-SIFT descriptor（descriptor mixed with RGL and CS-SIFT, RGL-CS-SIFT）. The results indicate that the RGL-CS-SIFT increases the matching speed and improves the ratio of image matching significantly, compared with SIFT and PCA-SIFT（Principal Component Analysis SIFT, PCA-SIFT） descriptors.

关 键 词：SIFT描述子 压缩感知 几何特征 图像匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]