Comparative competitive coding for personal identification by using finger vein and finger dorsal texture fusion

In this paper, we present a multimodal personal identification system that fuses finger vein and finger dorsal images at the feature level. First, we design an image acquisition device, which can synchronously capture finger vein and finger dorsal images. Also, a small dataset of the images has been established for algorithm testing and evaluation. Secondly, to utilize the intrinsic positional relationship between the finger veins and the finger dorsal, we perform a special registration on two kinds of images. Subsequently, the regions-of-interest (ROIs) of both kinds of images are extracted and normalized in both size and intensity. Thirdly, we develop a magnitude-preserved competitive code feature extraction method, which is utilized in both the finger vein and finger dorsal images. Furthermore, according to the preserved magnitude, a comparative competitive code (C2Code) is explored for finger vein and dorsal fusion at the feature level. The proposed feature map of C2Code, which contains new features of junction points and positions from the finger vein and finger dorsal image pairs, is extremely informative for identification. Finally, the C2Code feature map is fed into a nearest neighbor (NN) classifier to carry out personal authentication. Experimentally, we compare the performance of the proposed fusion strategy with that of state-of-the-art unimodal biometrics by using the established dataset, and it is found that there is higher identification accuracy and lower equal-error-rates (EERs).