In this paper, we propose a new clustered minutiae-based scale and rotation invariant fingerprint matching method. The major challenge faced in the existing latent fingerprint identification system is the lack of minutiae features in the fingerprint regions and hence there is a requirement to utilize the existing minutiae arrangements in the regions to identify the query fingerprint. We have clustered minutiae around a reference minutia and generated minutiae invariants to identify the fingerprint. In this paper, we propose two algorithms based on the minutiae neighborhood. To solve the geometrical constraints between the pairs of nearest points around a minutia, we propose the latent minutiae similarity (LMS) algorithm. Based on geometrical arrangements on the set of latent minutiae patterns around a minutia, we propose a clustered latent minutiae pattern (CLMP) algorithm. We test our algorithms on the FVC2004 and NIST SD27 criminal fingerprint databases. Proposed LMS, CLMP algorithms produced the highest 97.5% and 100% of Rank-1 identification accuracy respectively on plain FVC2004 dataset. Whereas, for NIST SD27 latent fingerprint database the proposed LMS, CLMP algorithms produced the highest Rank-1 identification accuracy of 88.8% and 93.80% respectively. Experimental results show significant improvement in the Rank-1 matching accuracy under random fingerprint scale and rotation condition compared to the state-of-the-art algorithms.

在本文中，我们提出了一种新的基于聚类细节的尺度和旋转不变指纹匹配方法。现有的潜在指纹识别系统面临的主要挑战是指纹区域中缺少细节特征，因此需要利用区域中的现有细节布置来识别查询指纹。我们在参考细节周围聚集了细节，并生成了细节不变以识别指纹。在本文中，我们提出了两种基于细节邻域的算法。为了解决细节附近的成对点之间的几何约束，我们提出了潜在的细节相似度（LMS）算法。根据围绕细节的一组潜在细节图案的几何排列，我们提出了一种群集的潜在细节模式（CLMP）算法。我们在FVC2004和NIST SD27犯罪指纹数据库上测试了我们的算法。提出的LMS，CLMP算法在普通FVC2004数据集上分别产生最高的Rank-1识别准确率97.5％和100％。而对于NIST SD27潜在指纹数据库建议的LMS，CLMP算法，最高的Rank-1识别准确率分别为88.8％和93.80％。实验结果表明，与最新算法相比，在随机指纹比例和旋转条件下，Rank-1匹配精度有了显着提高。在普通FVC2004数据集上，Rank-1识别精度分别为5％和100％。而对于NIST SD27潜在指纹数据库提出的LMS，CLMP算法，最高的Rank-1识别准确率分别为88.8％和93.80％。实验结果表明，与最新算法相比，在随机指纹比例和旋转条件下，Rank-1匹配精度有了显着提高。在普通FVC2004数据集上，Rank-1识别精度分别为5％和100％。而对于NIST SD27潜在指纹数据库建议的LMS，CLMP算法，最高的Rank-1识别准确率分别为88.8％和93.80％。实验结果表明，与最新算法相比，在随机指纹比例和旋转条件下，Rank-1匹配精度有了显着提高。