Hierarchical structure construction based on hyper-sphere granulation for finger-vein recognition

Abstract

Recently, the finger-vein (FV) trait has attracted substantial attentions for personal recognition in biometric community, and some FV-based biometric systems have been well developed in real applications. However, the recognition efficiency improvement over a large-scale database remains a big practical problem. In this paper, we propose an efficient and powerful hierarchical model based on hyper-sphere granular computing (HsGrC) for saving recognition cost. For a given FV database, samples are first viewed as atomic granules for building a basic hyper-sphere granule set. Using HsGrC, several different granule sets with multi-granularities are then generated by hyper-sphere granulation. To build a hierarchical structure of granule sets with granularity variation, a new quotient space relationship is established considering recognition efficiency improvement. Experimental results over a large finger-vein image database demonstrate that the proposed hierarchical model performs very well in computing cost reduction as well as recognition accuracy improvement.