Contactless finger recognition using invariants from higher order spectra of ridge orientation profiles

Abstract

A new method of biometric identity verification using images of fingers (contactless sensing) is presented. The method utilizes ridge orientation along lines between easily and reliably extracted key points and bispectral invariant features from the ridge orientation profiles. Rotation is corrected in the pre-processing stage after extraction of key points. Robustness to translation and scale are incorporated in the feature extraction. The method does not rely on minutiae extraction and has the potential for feature fusion from multiple fingers for improved performance. A radial basis function Support Vector Machine is trained to perform each identity verification. Results were obtained using 1341 index finger images from 41 individuals with 10-fold cross-validation. The system shows about 12% misses at a setting of 1% false alarms and the classification accuracy of the fused system is 92.99%. The performance can be improved with the use of multiple fingers. The the proposed methodology can facilitate high traffic, soft identity verification in busy premises such as shopping centers with the presentation of the hand as a person walks through.