A Gabor Filter-based Approach to Fingerprint Recognition

We propose a Gabor-Alter-based method for fingerprint recognition in this paper. The method makes use of Gabor filtering technologies and need only to do the core point detection before the feature extraction process without any other pre-processing steps such as smoothing, binarization, thinning, and minutiae detection. The proposed Gabor-filter-based features play a central role in the processes of fingerprint recognition, including local ridge orientation, core point detection, and feature extraction. Experimental results show that the recognition rate of the k-nearest neighbor classifier using the proposed features is 97.2% for a small-scale fingerprint database, and thus that the proposed method is an efficient and reliable approach. INTRODUCTION Because of unchangeability and uniqueness, fingerprints have been widely applied in several fields of personal identification such as criminal investigation, access control, and Internet authentication. In fingerprints, the ridge structures can provide global and local information. Therefore, most researchers use the local ridge orientation of fingerprints for classification and use the minutiae, a group of ridge endings and bifurcations, for identification. However, the minutiae-based approach is very sensitive to noise and deformation. For instance, false ridge endings and bifurcations may appear due to blurred or overinked problem. Moreover, ridge endings and bifurcations may disappear because the finger is pressed too hard or too light [l]. In other words, the performance of minutiae extraction algorithms relies heavily on the quality of input images. Furthermore, the main steps for a minutiaebased approach are image acquisition, enhancement, ridge extraction, thinning, minutiae detection, and matching [2] [3]. For a small-scale fingerprint recognition system, however, it would not be efficient to process all the steps and the recognition result will be heavily dependent on the accuracy of each step. 0-7803-5650-0/99/$10.00 8 1999 IEEE 37 1 The Gabor-filter-based features, directly extracted from gray-level fingerprint images, have been successfully and widely applied to texture segmentation [4] [ 5 ] , face recognition [6], and handwriting recognition [7]. In fingerprint applications, the Gabor-filter-based features for enhancement [8], classification [9] and recognition [lo] are also proposed. The characteristics of the Gabor filter, especially the frequency and orientation representations, are similar to those of the human visual system [ll]. In this paper, we will propose the use of Gabor filter-based features for all steps of fingerprint recognition consisting of local ridge orientation, core point detection, and recognition. At last, we will use a small-scale fingerprint database to test the performance of the proposed method. GABOR FEATURES AND THE PROPOSED METHOD In [7], the general form of a 2D Gabor filter is defined by where xe, = x cos & + y sin e k and ye, = x sin ek + y cos @ k , f is the frequency of the sinusoidal plane wave, d k is the orientation of the Gabor filter, and 5, and f l y are the standard deviations of the Gaussian envelope along the z and y axes, respectively. To analyze the Gabor filter in terms of the evensymmetric and the odd-symmetric, we express eqn. 1 in the complex form h = he,,,, + i h o d d , where Since most local ridge structures of fingerprints come with well-defined local frequency and orientation, f can be set by the reciprocal of the average interridge distance and m as the number of orientations for calculating & = n(k l)/m, k = 1,. . . , m. Also, the cosine/sine form and the sinusoidal-shape of the Gabor filter is suitable for modeling ridge structures and smoothing out noise, respectively. After deciding the parameters of the Gabor filter, the magnitude Gabor feature, the even one, and the odd one at sampling point (X, Y) can be defined as follows: