A Study of Various Classification Algorithms to Reveal Anomaly during Fingerprint Presentation on Biometric Sensor Based on Naïve Feature Vector On Two Separate Region

Fingerprint biometric-based application suffers from a problem, in which no genuine biometric sample known as a spoof sample is classified as a live presentation. To overcome this problem the research community is proposing new algorithms or adopting existing state-of-art. In continuation, this work proposes a naïve feature vector computed from two independent image regions to analyze various classifiers' performance to validate feature vector design. In this work, the first step is to calculate the region of interest (ROI) using factorized directional band-pass (FBD) segmentation algorithm to remove unwanted regions. The two regions proposed are region-I and region-II. The (region-is composed of four (non-overlapping) sub-regions calculated using the Otsu thresholding algorithm. The second region (region-comprises six sub-regions (overlapping) using activity mask, noticeable artifacts mask, noise mask, and their image negatives. For feature selection, this work proposes to use five different no-reference image quality assessments (IQA) metrics. These are perception-based image quality evaluator (PIQE), naturalness image quality evaluator (NIQE), and its model blind/reference less image spatial quality evaluator (BRISQUE) and its model. The two feature design consists of a total of 38 and 24 features respectively. To test the accuracy of the proposed methodology all 25 binary classifier algorithm available in MATLAB has experimented with 10-fold cross-validation. This experiment uses cross-sensor and cross-material experimental validation on two different datasets LivDet2013 and LivDet2015. The proposed method achieved 99.8% reasonable accuracy with the Bagged Tree Ensemble learning classifier with cross-sensor and cross-material combinations using 38 features. Further, another classifier performs fairly well with 24 features in a lower combination of datasets which is comparable with existing state-of-art methods. The experimental result verifies the effectiveness of the proposed method, which is a good candidate for a futuristic open-set solution. This experiment can be recreated and experimented on a larger combinational dataset for the generalization of results as future work for the proposed study.