Fingerprint Genuinity Classification

Abstract

AbstractA dataset that consists of both contactless and contact-based fingerprints was collected using various instruments like sensors, images, etc. Sensor Interoperability issue was handled using this system. Implementation of contactless fingerprint technologies widely depends on the progressive capability to match contactless 2D fingerprints with contact-based fingerprint databases. Contactless 2D fingerprint identification is more safe and it allows deformation-free imaging for advanced accuracy. Here, an Auto Stack Encoder for accurately matching contact-based and contactless fingerprint images was proposed. Auto Stack Encoder is a neural networks for which the input is similar as the output as they work by reducing the input into a latent-space representation and then again constructing the output from the corresponding image. This Auto Stack Encoder is trained using the images of fingerprint and features and classified as genuine (1) or fake (0). Therefore, an additional robust thin-plate spline model from the concatenation of deep extracted feature vectors made from different networks. This paper shows the Unsupervised learning approach for the Fingerprint Genuinity Classification of Contactless and Contact based fingerprints using the Auto Stack Encoder.KeywordsFingerprint classificationPreprocessingFeature extractionRobust thin plate spline modelDeformation correction modelAuto stack encoder