Novel invariant feature encoding method for finger-vein IR images

The feature representing method based on encoding has advantages in illumination invariance, calculation efficiency and feature representation. It is one of the new feature extracting methods. The finger-vein images were captured in a perspective view by using infrared light source. Due to the complex biological tissue and the imaging mode, the image quality was usually poor. The Gabor filtering was used to enhance the image texture and the local graph structure encoding was adopted. A new feature encoding method was proposed with emphasis on a new local graph structure which was symmetric and crossed in neighborhood. The image texture of local neighborhood was converted to weighted encoding strings. The filtered image of each channel of Gabor filter was encoded in different directions so as to express the information of position and gradient in the neighborhood adequately, and it was rotation invariant as a result. The experiments results show that new method achieves better performance in finger recognition than common feature encoding methods and is more robust in rotation.