Point grouping method for finger vein recognition

Abstract

In finger vein recognition, vein points-based methods classified the image points into vein points and non-vein points and only measured the vein points for the recognition. All points-based methods utilized the features of all points regardless of the difference of vein points and non-vein points. This paper proposes a point grouping method for finger vein recognition by integrating the advantages of the above-mentioned two kinds of methods. In the proposed method, all image points are considered in the recognition, and the points are classified into multiple groups in both feature extraction and similarity measurement. The matched (unmatched) points are separately found from each group pair of the enrolled image and the probe and are fused to obtain the similarity (dissimilarity) score. Moreover, we incorporate the idea of the point grouping into two popular finger vein recognition methods and devise the corresponding point grouping assisted methods, namely, point grouping assisted anatomy structure analysis-based vein extraction and point grouping assisted Gabor. The experimental results on finger vein databases demonstrate the efficacy of the point grouping assisted methods. The equal error rate values of our point grouping methods are lower about 1% than their base methods on SDUD in the multiple templates' mode and on HKPUD in the single template mode.