A Novel Feature Selection Method Based on Bio-inspired Artificial Intelligence for The Feature Selection of Bio-metric Data

Abstract

Recently, both industry and academics are paying increasingly more attention to bio-metric authentication using palm vein patterns due to the benefits it provides over traditional methods, like those that rely on a fingerprint, iris, or face. However, feature extraction of the palm vein feature, which seeks to collect the most distinctive features that uniquely identify individuals might have redundant and useless features, which would reduce the performance of the bio-metric system and cause a dimension issue. This paper presents an improved feature selection technique based on a bio-inspired artificial intelligence algorithm called Discrete Artificial Bee Colony (DABC) for palm vein bio-metric authen-tication. The best set of features from those offered by Gabor feature extractor are selected using the DABC algorithm. In attempt to grant additional diversity and effectively explore the solution space, a variant of Opposition Based Learning (OBL) notion is implemented. The proposed approach, termed Modified Opposition based DABC (MODABC), has been evaluated using the CASIA palm vein database in comparison to a number of feature selection algorithms based on bio-inspired artificial intelligence. The experimental findings demonstrate that the presented method perform better that the literature in a number of areas, including accuracy, FRR, FAR, feature vector size, and fitness values. The method being proposed enables providing a suitable balance between the feature vector size and the authentication rate.