一种有效的全景图拼合预处理算法  

An Efficient Preprocessing Algorithm to Panoramic Image Stitching

作　　者：张桦[1] 孙济州[1] 谭磊[2] 

机构地区：[1]天津大学计算机科学与技术学院,天津300072 [2]天津理工大学计算机科学与技术学院,天津300191

出　　处：《光电子．激光》2007年第6期746-749,共4页Journal of Optoelectronics·Laser

基　　金：天津自然科学基金资助项目(033600211);天津教委基金资助项目(2004BA09)

摘　　要：全景图拼合的预处理主要包括对样本图像的特征提取以及相邻图像的特征匹配。本文首先改进Harris角点检测算子,以便准确提取样本图像的特征点并赋予特征描述符;提出一种基于小波系数的特征索引算法,实现了图像特征点对的快速搜索和两幅图像之间的匹配。实验结果表明,该算法得到的匹配点精确,搜索效率高,能够实现全景图的无缝拼接。The preprocessing to panoramic stitching mainly includes extracting features form sample images and matching. Then to images of overlapping. In this approach,an extended Harris corner detector is proposed by us to accurately extract feature points from an image and to assign a feature descriptor for each of them. After that,we suggests a new feature indexing algorithm based on wavelet coefficients which enables comparing features in neighboring images more efficiently. Experiments show that feature matches are accurate and efficient,and seamless panoramic image stitching can be produced.

关 键 词：特征点 小波变换 K-近邻 图像拼接 

分 类 号：TN911.73[电子电信—通信与信息系统]