Application Of Reliable Correspondence Propagation Algorithm For Fingerprint Cross Matching

Fingerprint matching algorithm is the core part of the Automatic FingerprintIdentification System and has an important impact on system performance. Commonlyused matching algorithms are feature-based and image-based. Since the image-basedmatching algorithm usually has higher rejection rate and error rate, it was used a few.While the feature-based algorithm usually has accurate calculation result on goodquality fingerprint, it used more generally. However, the traditional feature-basedfingerprint matching algorithm usually encounter such an issue，that the top few reliablecorrespondences are often easy to obtain,while as the desisred correspondece numberincreases,the overall matching accuracy may fall drastically. In addition, thematching of two fingerprints come from different acquisition instruments, due tothe acquisition principle differences between the sensors, the two fingerprints will besuffer a relatively large deformation, thus affecting the performance of the algorithm.In order to address the above issues, we apply the Reliable CorrespondencePropagation (RCP) algorithm to fingerprint recognition, and obtain good results. Thisalgorithm mainly consists of two steps:(1) Determine the initial correspondencebetween features.(2) Propagate the initial correspondence in spatial space and featurespace. First, a detailed description of the theory of the RCP algorithm is given, and aclosed-form solution is deduced. Then, we combine it with Tico Descriptor and domatching experiment on database FVC2004DB1. The results shows that theperformance of the RCP algorithm is better than single alignment method but bad thanmultiple alignment method. However, after the improvement of the initialcorrespondence, the algorithm get a equivalent performance with the multiple alignmentmethod. Finally, we apply the RCP algorithm to the cross matching of fingerprints, andtest it between different databases. The results shows that the RCP algorithm can alsoget a desired performance on cross matching. Besides this, compared with thetraditional fingerprint matching algorithms, our algorithm can effectively solve thedeformation problem in cross matching.