A Novel Approach for Fingerprint Classification using Optimized Support Vector Machine

Individuals can be identified using biometric approaches. With the development of information technology biometric authentication is widely used. Fingerprint is one of the promising authentication technique used. In fingerprint authentication the fingerprint image is processed to extract personalized features and further the extracted features pass through a classification technique. In the classification process takes longer time with large number of users. Whereas, if we categories the users to different category then this classification process becomes much faster because, the individual classification is performed only inside the similar category. Since the search space is reduced to only one category of fingerprints, so the process becomes much faster. In general the fingerprints are classified into five categories: Arch (A), Left loop (L), Right loop (R), Tented Arch(T), Whorl (W). These classification systems are highly dependent on a preprocessing phase that increases the processing time at test stage. Motivated by the recent success of deep learning techniques in many computer vision tasks, the proposed system in this paper consists of a pre-trained Support Vector Machine (SVM) fine-tuned on a subset of the NIST SD4 benchmark database. The SVM architecture contains a VGG16 model. The main contribution of this paper is to show the performance of transfer learning to the problem of fingerprint classification.