English

应用科学学报

• 论文 • 上一篇    下一篇

自动图像拼接中的一种特征提取和匹配方法

张静, 严壮志，邵世杰，王牧云，王黎明
  

  1. 上海大学 通信与信息工程学院，上海200072
  • 收稿日期:2007-10-10 修回日期:2007-12-20 出版日期:2008-05-31 发布日期:2008-05-31

Improved Method of Feature Extraction and Matching for Image Mosaic

ZHANG Jing, YAN Zhuang-zhi, SHAO Shi-jie, WANG Mu-yun, WANG Li-ming
  

  1. School of Communication and Information Engineering, Shanghai University,Shanghai 200072, China
  • Received:2007-10-10 Revised:2007-12-20 Online:2008-05-31 Published:2008-05-31

被引次数

30

摘要： 在比较目前特征提取和匹配的几种方法比较的基础上，提出了一种基于改进特征提取和匹配的拼接方法，使得图像拼接的质量和速度得到提高。该算法首先利用改进的尺度不变特征变换（scale invariant feature transform，SIFT）特征提取方法获得图像特征点，其次利用近似最近邻匹配进行特征匹配并引入随机抽样一致性（random sample consensus，RANSAC）算法去除误匹配对，最后根据匹配的特征点对得到图像间的变换参数进行拼接和融合。该算法具有很强的鲁棒性，允许图像有缩放变换、旋转变换，不受图像噪声、色差的影响。经实验证明，该方法可实现高质量快速的拼接系统。

关键词: 图像拼接, 尺度不变特征变换, 近似最近邻匹配, 随机抽样一致性

Abstract: By comparing with some common approached of feature extraction and matching, the paper proposes an improved method of feature extraction and matching for image mosaic which improves image quality and processing speed. It uses scale invariant feature transform (SIFT) to extract invariant features from images, approximate nearest neighbor searching and random sample consensus (RANSAC) to perform reliable matching. Parameters of the transformation between images are obtained from the matched feature points to realize image stitching and blending. The feature points are invariant to affine transformation, noise contamination and illumination variation, leading to robustness of the method. Experimental results show that the proposed method is fast and can produce high quality image mosaic.

Key words:

image mosaic, scale invariant feature transform , approximate nearest neighbor matching, random sample consensus (RANSAC)