FINGERPRINT RECOGNITION USING WAVELET FEATURES Marius Tico, Eero Immonen, Pauli Ramo, Pauli Kuosmanen, and Jukka Saarinen Digital Media Institute, Tampere University of Technology P.O.BOX 553, "-33101, Tampere, FINLAND E-mail: tic0 @cs. tut.fi ABSTRACT A new method of fingerprint recognition based on features extracted from the wavelet transform of the discrete image is introduced. The wavelet features are extracted directly from the gray-scale fingerprint image with no pre-processing (i.e. image enhancement, directional filtering, ridge segmenta- tion, ridge thinning and minutiae extraction). The proposed method has been tested on a small fingerprint database us- ing the k-nearest neighbor (k-NN) classifier. The very high recognition rates achieved show that the proposed method may constitutes an efficient solution for a small-scale fin- gerprint recognition system. 1. INTRODUCTION The use of fingerprints for personal identification has a very long history. Archaeological evidences reveal that finger- prints have been used as a form of identification since 7000 BC [I]. The formation of the fingerprints depends on the initial conditions of the embryonic development, and their ridge pattem is unchanged throughout the entire life (im- mutability). In addition, since the first scientific studies of fingerprints in the mid-1800 till today no two fingerprints from different fingers have been found to have the same ridge pattem (uniqueness). Both the immutability and the uniqueness properties have determined the use of finger- print matching as one of the most reliable techniques of people identification [2]. A fingerprint image exhibits a quasiperiodic structure of ridges (darker regions) and val- leys (lighter regions). The local characteristics of this stmc- ture (i.e. ridge endings and ridge bifurcations) called minu- tiae form a unique pattem for each fingerprint. Various approaches of automatic fingerprint matching have been proposed in the literature. They include minutiae- based approaches, and image-based approaches as the most prominent classes of fingerprint matching methods. Minutia- based approaches are the most popular ones being included in almost all contemporary fingerprint identification and ver- ification systems. Although rather different from one other the minutiae-based approaches require extensive preprocess- ing operations in order to reliably extract the minutia fea- tures. The preprocessing operations include image enhance- ment, orientation flow estimation, ridge segmentation, ridge thinning, minutiae detection [3-71. In addition, a minu- tiae purification stage is also required in order to reduce the number of false minutiae erroneously detected in noisy fingerprint images [8, 93. The minutiae form a pattem of points, and hence fingerprint matching may be seen as a point pattem matching problem. Several minutiae attributes (e.g., position, type, orientation, ridge count) are used in order to simplify the general point pattem matching prob- lem which is essentially intractable [4-71. Image-based ap- proaches do not use the minutiae features for fingerprint matching. They are usually applied directly onto the gray- scale fingerprint image without pre-processing. and hence they may achieve higher computational efficiency than minu- tiae-based methods. In addition, the image-based approaches may be the only choice to match fingerprints which have too low image quality to allow a reliable minutiae extraction. The main disadvantage of image-based approaches consist in their limited ability to track with variations in position, scale and orientation angle. Usually the variation in po- sition between the two fingerprints is cancelled by choos- ing a reference point in each fingerprint. Such reference point may be the core point which can be detected using for example methods like those proposed in [lo-171. Image- based approaches include methods based on optical correla- tion [13, 141 and transform based features [15]. In this paper we propose an image-based method of fin- gerprint recognition. The fingerprint pattems are matched based on wavelet domain features which are directly ex- tracted from the gray-scale fingerprint image without pre- processing. The paper is organized as follows. This introduction serves as the first section. The following section presents the proposed method of wavelet features extraction. Exper- imental results including k-NN classification experiments are shown in Section 3. and finally a brief conclusions sec- tion will summarized the paper. 11-2 1 0-7803-6685-9/01/$10.0002001 IEEE