Contactless multiple finger Segments based Identity verification using information fusion from higher order spectral invariants

Abstract

A methodology for identity verification from contactless finger images using ridge orientation profiles and spectral invariant features is extended to use fusion at data, feature levels, and a combination of both levels using multiple finger segments. Performance is evaluated on 1341 images (selected 24 Megapixel video frames obtained with the finger-to-camera distance between 12 and 20cms, reduced to finger regions of about 1250 x 3000 pixels) from 41 individuals of different ethnicities. Features are extracted from profiles along key lines between landmarks that facilitate segmentation. The methodology is designed by means of the segmentation procedure, the invariant features, and fusion techniques to be robust to geometric and photometric transformations and partial occlusion. Feature fusion, data fusion, and a combination of the two are tested. The methodology yields around 5% EER with a combination of data and feature fusion providing the best performance. It can be applied to soft, on-the-move biometric systems.