摘要
针对实时匹配的要求,提出一种基于DOG特征点的快速图像匹配算法,用以匹配只存在平移和较小旋转的序列图像。该算法通过求高斯差分算子在尺度空间上的局部极大值和极小值提取特征点,然后根据圆旋转不变特性生成20维的旋转不变特征描述子,并充分利用特征点的区域特征和灰度特征进行匹配,最后根据序列图像对应特征点之间的距离基本保持不变的特性剔除错误的匹配点。实验结果表明该算法快速有效,而且对噪声影响不敏感,具有很强的实用性。
In order to match in real-time,a fast algorithm based on DOG feature points is presented in this paper, which can be used to find corresponding points between serial images in the case of translation and minor rotation. Feature points are extracted in the process of searching for the local maximum and minimum of difference of Gaussian function in scale space. Then rotation invariant feature descriptors of 20 dimensions are generated based on that the shape of circle kept same when rotated. Subsequently,matching operations are performed by making full use of the location and gray-value of feature points. At last,the false matching points are removed according to the characteristic that the distance of corresponding points maintained invariant between serial images. The algorithm is proved to be fast, robust to noise,effective and practical by the experiment results.
出处
《现代电子技术》
2008年第4期128-130,136,共4页
Modern Electronics Technique
基金
国家高技术研究发展计划(863计划)课题(2006AA801412)
关键词
图像匹配
特征描述子
尺度空间
高斯差分算子
image matching
feature descriptor
scale space
difference of Gaussian
