基于SIFT联合描述子的航拍视频图像镶嵌  

A Better Aerial Video Image Mosaic Algorithm Using SIFT Combined Descriptors

作　　者：程咏梅[1] 张绍武[1] 花永强[1] 柳丹[1] 雷宝权[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《西北工业大学学报》2010年第1期51-56,共6页Journal of Northwestern Polytechnical University

基　　金：国家自然基金重点项目(60634030);航空科学基金(2007ZC53037)资助

摘　　要：针对存在明显旋转、尺度或模糊等变化的图像,文章提出了一种基于SIFT联合描述子的航拍视频图像镶嵌算法。该算法首先计算经典的SIFT特征局部描述子,然后通过引入图像的形状信息得到每个特征点的全局描述子,进而获得联合描述子;对联合描述子中的局部描述子采用欧氏距离、全局描述子采用卡方距离进行双向匹配,通过加权平均法融合得到联合描述子的匹配距离;采用RANSAC算法精确估计仿射变换模型参数。实验结果表明:该算法具有较好的鲁棒性,可获得较好的镶嵌结果。Aim. The introduction of the full paper points out that： （1） Ref. 1 uses local descriptors; （2） Refs. 1 and 2, both by D. G. Lowe, mention the possibility of utilizing the similarity between consecutive frames to obtain global shape information. We believe that we have succeeded in utilizing the above-mentioned similarity to obtain actually global shape information. Section 1 explains the SIFT （ scale invariant feature transform） combined descriptor; its core consists of： it firstly calculates the classical features of local SIFT descriptor, receives the global descriptor of each feature point by introduci,lg the shape ilfformation of the image, and then obtains combined descriptors. Fig. 3 in section 2 presents the flow chart of aerial video image mosaic algorithm using SIFT combined descriptors. Section 3 explains image matcbing; its core consists of： for bidirectional matching, Euclidean distance is.utilized for local descriptor, Chi-square distance for the global one, and then, the matched distance of combined descriptors is obtained with the weighted average integration method. Section 4 explains the parameter estimation obtained with the RANSAC （random sample consensus） algorithm and the least squares method. The experimental results, presented in Figs. 5, 6 and 7, show preliminarily that our algorithm has better robustness and is indeed better than that of Ref. 1.

关 键 词：航拍视频图像 镶嵌 SIFT联合描述子 双向匹配 仿射变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]