On accurate orientation extraction and appropriate distance measure for low-resolution palmprint recognition

Orientation feature has been demonstrated to be one of the most effective features for low resolution palmprint recognition. In this paper, using steerable filter, we investigate the accurate orientation extraction and appropriate distance measure problems for effective palmprint recognition. First, we use high order steerable filter to extract accurate continuous orientation, and quantify it into discrete representation. Then, for effective matching of accurate orientations, we propose a generalized orientation distance measure. We further extend the distance measure for matching of discrete orientations, and show that several existing distance measures can be viewed as its special cases. Experimental results on both Hong Kong PolyU and CASIA palmprint databases show that the proposed method can obtain state-of-the-art verification accuracy. With the support of a look up table, the proposed method also enables small template size and satisfactory matching speed for practical applications.