Fingerprint Enhancement with Dyadic Scale-Space Jiangang Cheng, Jie Tian, Tanghui Zhang (AILAB, Institute of Automation, Chinese Academy of Science, Beijing, 100080) e_mail: tian@doctor.com Abstract Fingerprint enhancement is a critical step in finger- print identification. Most of the existing enhancement uses a set of contextual filters to enhance fingerprint. The main drawback of these methods is these contextual filters based on the local information of the fingerprint, such as ridge width, orientation, curvature et al. These informa- tion are unreliable in the areas corrupted by the noise. This paper introduces the scale space theory in the com- puter vision to enhance the fingerprint. In the enhance- ment process, decompose fingerprint into a series of im- ages and organize the images by finer to coarser scheme. Thus a globe and integrate interpretation is available and it enable us to get rid of the influence of noise to the larg- est extent. Experiments show our algorithm is fast and has excellent performance. Supported by Natianal Natural Science of Fundation of China Undergranted Nos: 60072007, 69931010 1. Introduction Most of the existing enhancements use a set of con- textual filters to enhance the fingerprint based on the local estimates, such as ridge width, orientation, and curvature et al. Due to the existence of noise, these methods result in unreliable estimates and sometimes introduce artifacts. The method of Lin Hong et al.[1] uses Gabor filter to enhance the fingerprint image. The main drawback of this method lies in the fact that false estimate of local ridge direction will lead to poor enhancement. In literature[2]， Andrés et al. treat the ridge in a local region as a cylindri- cal sine wave. This module is unsuitable in areas of the singular point, because the ridge curls sharply. Due to the existence of noise, some fingerprint char- acters are submerged. False estimates will lead to poor enhancement. When enhancing fingerprint, we utilized the whole information to get rid of the noise and reserve the true characters. This paper introduces the dyadic scale-space (DSS) in the computer vision to enhance the fingerprint. We first decompose the fingerprint image into a series of images in different scales, then analyze and organize whole charac- ters and details, at last combine the creditable information to enhance the fingerprint image. The steps of our enhancement are shown in figure 1 Section 2 introduces the scale space theory. Section 3 we use DSS theory to enhance the fingerprint. And sec- tion 4 gives our experiments, we conclude in the last section Input fingerprint image Preprocess with DSS Compute orientation Binarize fingerprint image Compute mean ridge width Output enhanced image No Yes k=k+1 W k 2 log < Figure 1. Block diagram of the enhancement algorithm 2. Linear Scale Space Scale-space theory [3] provided a canonical framework for modeling visual operations at multiple scales. The scale-space representation : L 2 R × R → R of a two- dimensional image is defined as the one-parameters family of functions obtained by convolving with Gaus- sian kernel: f f 1051-4651/02 $17.00 (c) 2002 IEEE