A Comparative Study of Combining Multiple Features for Fingerprint Recognition

A number of different features, besides minutiae, have been used for fingerprint matching. There include orientation-based minutia descriptor, FingerCode, ridge feature map, orientation map, and density map. Previous studies have shown that the performance of a fingerprint recognition system can be improved by combining these features with minutiae through a fusion strategy. In this paper, a comparative study of combining multiple features is carried out on a large database. Based on the experimental results, some useful conclusions can be drawn: (1) A larger improvement can usually be obtained as we combine more features; (2) However, beyond 4 features, the performance improvement is very small. (3) MD (minutiae and orientation-based minutia descriptor) are the best of two-feature combinations. It is interesting that the single feature D (minutia descriptor) does not perform the best, but the combination of MD (minutiae and minutia descriptor) performs very well. (4) Although the combinations of MO and MD performs the best among the two-feature combinations, MDO performs not so good as MDR (minutiae, minutia descriptor, and ridge density map) due to the correlativeness between D and O. These observations are helpful for designing practical fingerprint matching systems.