Abstract: Most of the fingerprint matching algorithms used in automatic fingerprint identification systems (AFIS) are based on the accurate extraction of minutiae. However, when these methods are used to deal with highly distorted, rotational, and usually partial fingerprints collected from crime screen, their performance drops significantly. On the basis of similar triangle matching algorithm, this paper conquers scale problem by finding the positional relationship between SIFT (scale invariant feature transform) feature and second level feature. In addition, this paper also proposes a Bayesian statistical inference method to fuse the two kinds of algorithms, which is named as similar triangle SIFT feature (STSF) algorithm. The experimental results show that STSF algorithm effectively increases the precision and efficiency of partial fingerprints matching.

Key words: fingerprint matching, feature fusion, partial fingerprint

摘要： 目前大部分指纹自动识别系统（automatic fingerprint identification systems，AFIS）所采用的特征点匹配算法需以准确提取特征点为前提，这些算法在面对存在高度畸变且残缺不全的现场指纹时，往往难以准确识别指纹图像。在相似三角形匹配算法的基础上，研究了SIFT（scale invariant feature transform）特征点与二级特征点之间的位置关系，克服了相似三角形之间尺度不一的问题。此外，提出了一种基于贝叶斯统计推断的相似三角形与SIFT融合算法（similar triangle SIFT feature，STSF）。实验结果表明，STSF算法能够有效提升残缺指纹匹配的精度和计算效率。

关键词: 指纹匹配, 特征融合, 残缺指纹