摘要： 当不同成像条件下图像局部内容的特征存在明显差异时，会导致特征描述子难以正确匹配。为此，提出一种基于稳定区域的图像特征描述子。利用最大稳定极值区域检测算法提取图像的稳定区域，结合图像特征点的位置和尺度信息，根据稳定区域计算特征主方向，在适应于特征点所属稳定区域尺度的较大邻域内，生成基于对数-极坐标系的特征描述子。应用结果表明，该描述子在图像局部内容变化较大时仍具有旋转不变性和稳定的匹配性能，能有效解决印刷电路板CT图像的配准问题。

关键词: 特征描述子, 最大稳定极值区域, 印刷电路板图像, CT图像配准, 尺度不变特征变换, 相似区域

Abstract: Aiming at the large difference of local image context under different imaging condition, which leads to descriptors can not be matched correctly, this paper presents a descriptor based on stable region. It uses Maximally Stable Extremal Region(MSER) detector to extract the stable image region, then assigns the consistent orientation based on stable regions and computing the descriptor based on log-polar coordinate in the larger neighborhood that adapted to keypoint’s scale. It applies the descriptor in the CT image registration of Printed Circuit Board(PCB). Application results show that this descriptor is invariant to image rotation and can be used to perform reliable matching when the local image context changes, and has the distinctive performance to resolve ambiguities. The descriptor can work well in the CT image registration of PCB.

Key words: feature descriptor, Maximally Stable Extremal Region(MSER), Printed Circuit Board(PCB) image, CT image registration, Scale Invariant Feature Transform(SIFT), similar region

中图分类号: 

• TN911.73