A Personal Identification Method Based on the Geometrical Patterns of Veins Grid in a Finger

As a new biometric technique, finger recognition has attracted lots of attentions and used in wide range of applications. Finger vein recognition is a physiological characteristics-based biometric technique; it uses vein patterns of human finger to perform identity authentication. Vein patters are network of blood vessels under a person’s skin. Even in the case of identical twins the finger-vein patterns are believed to be quite unique. This makes finger vein detection a secure biometric for individual identification. In this paper a feature set based on the local binary projections of veins grid body is proposed to be used for personal identification. The proposed system consists of three main stages, which are: preprocessing, feature extraction, and matching. Since near infrared NIR vein images suffer from low contrast, and low noise; which make the extraction task of accurate veins grid become hard. For this reason a sophisticated preprocessing process needs to be accomplished to ensure high identification rates. The applied steps in preprocessing stage are: histogram equalization (to improve the contrast of the image). Also the brightness compensation step is applied to suppress the background and to make grid body more visible, and to make the segmentation task easier. Finally two levels of thinning are applied to make the grid appearance more localized.

Due to fast implementation, and both rotation and scale invariant features requirements; a feature set based on the local binary projection (in four directions: vertical, horizontal, main diagonal, and second diagonal) is adopted. The geometrical moments are calculated for the four direction projections which represent the discriminating local finger vein features.

The developed system was tested over SDUMLA-HMT finger-vein database collected from 106 volunteers using their index, middle and ring fingers of both hands. The collection for each of the 6 fingers is repeated for 6 times to obtain 6 finger vein images. The test results indicated that the equal error rate of our proposed is 99.2%. Increasing the number of learning sample leads to improvement the identification rate up to (100%).