1:N fingerprint classification
Biometric recognition systems are widely used to recognize an individual. Fingerprints is a biometric identifier and are today widely used in smartphones for biometric recognition. The fingerprint software used in smartphones are great and fast, and usually implemented for one person usage. A fingerprint software used for smartphones often conducts a one-to-one comparison between the sample and the enrolled templates. Software systems that can be used to recognize a person out of many is desirable. Such a system would conduct one-to-many comparisons. However, using the existing software in smartphones in a product used by many people would be too slow. As it has to conduct many one-to-one comparisons between the sample and the enrolled templates.
This thesis examines whether it is possible to classify the enrolled templates by K-centroid, such that the number of comparisons in the one-to-many authentication problem is reduced. It was shown that the bag-of-words representation of an image is the best feature to use, together with the cosine similarity, during classification. To further improve the clustering, SVM and balancing were also applied. This combination showed good results and is the most promising method out of all methods examined in this thesis.