This paper presents a feature-level fusion system for multimodal personal authentication by using finger vein and finger
dorsal images. A user-friendly imaging device for multimodal biometric identification is designed. Based on this device, we
have established a finger vein and finger dorsal image database [55]. In considering the inherent positional relationship between
finger veins and finger dorsal texture, an ROI extraction algorithm is proposed, which utilizes the positional information
of finger vein images to extract the corresponding ROI in the finger dorsal images. In that the two kinds of images
contain ample line features in the vertical and horizontal directions respectively, we investigate the junction points for fusion.
Therefore, a feature level fusion strategy is proposed to make use of the significant orientation features of the two
modalities, discard the less discriminative ones, and add positional information of the fusion features, i.e. junction points.
Experimentally, higher recognition accuracy and lower EERs are achieved in comparison to other multimodal and unimodal
identification methods, which demonstrate that the attempt in using the intrinsic relationship of texture on both the upper
and lower sides of the finger is efficient and promising for multimodal biometrics. Continuous efforts are being made to expand
our database and improve the performance in terms of finger rotation. The exploring of other fusion schemes that utilize
the unique positional relationship of the finger veins and finger dorsal for recognition will be our future work.