Hybrid CNN-Ensemble based Classifier for Touchless Fingerprint Classification

The outstand ascent in software computing has expanded the skyline for various applications to make our life easy in decision making. Among the significant requests, security frameworks have consistently been the predominant one to guarantee authenticity. Fingerprint classification techniques have acquired wide-spread consideration for personalized authentication. Automated fingerprint classification is one of the mostly used human identity verification system. The touchless fingerprint identification and classification system offers a higher user convenience and hygiene compared to conventional touch-based identification especially, during this pandemic situation. Recently, convolutional neural networks are trained to achieve better performance on fingerprint classification. Considering the scope for finger impression classification framework, in this paper a novel and vigorous ensemble based touchless finger impression classification system is proposed. By and by, it is extremely regular that the performance of a trained solitary classifier is fluctuated on various datasets, which demonstrates that a similar learning method may train solid classifiers on some datasets, yet frail classifiers might be prepared on other datasets. It is additionally conceivable that a similar classifier shows extraordinary execution on various test sets. To resolve the above issue, development of multilevel ensemble approaches have been very imperative to work on the overall execution and make the execution predictable on various datasets. The aim of this paper is to develop a structure that includes CNN-based feature extraction from the dataset and mathematical combination of multiple classifiers trained on different feature sets, which are set up through correlation-based filter feature selection approach applied to the original feature set extracted utilizing CNN. The test results obtained by conducting experiments on PolyU 3D fingerprint dataset show that the fusion of classifiers can accomplish good classification accuracy.