Finger vein indexing based on binary features

Finger vein indexing refers to the process of creating clusters of finger vein samples based on the extracted features from the image. In large scale finger vein identification, the input probe image needs to be compared with a large set of gallery images to match the identity. Clusters can be used to confine the process of identity match of the images present in the gallery to the probe sample of the same cluster. In this work, we explored a new finger vein indexing and retrieval scheme with K-means clustering and pre-selection of features & comparison(PSFC). For a input probe sample, the centroid with the smallest distance will be chosen and compare with them for identification with K-means and P% of features (First, Last and Random) are compared with gallery samples features to choose for identification with PSFC. Extensive set off experiments are carried on a large scale data set of 2850 unique instances created using publicly available finger vein databases. Single cluster search with K-means clustering demonstrated the performance with pre-selection error of 9.38% (hit rate of 90.62%), Multi-cluster search with K-means clustering has achieved best performance with pre-selection error of 2.53% (hit rate of 97.47%) and PSFC has demonstrated the efficiency with pre-selection error of 9.58/9.85/8.05% (hit rate of 90.42/90.15/91.95%)for first/last/random features.