Modern Multiresolution Techniques for Fingerprint Recognition

Abstract

Using biometrics in recognition of persons has received more and more attention in the last years, due to the necessity to improve the information security and access restrictions of authentication systems. Fingerprint is considered the most practical biometrics due to some specific features which make them widely accepted. Reliable feature extraction from poor quality fingerprint images is still the most challenging problem in fingerprint recognition system. So it needs a lot of pre-processing steps to improve the quality of fingerprint images, then it needs a reliable feature extractors to extract some distinctive features. Recently, multiresolution transforms techniques have been widely used as a feature extractor in the field of biometric recognition. These features can be used as an identification marks in fingerprint recognition. The goal of this thesis is to develop a complete and an efficient fingerprint recognition system that can deal with poor quality fingerprint images. To deal with poor quality fingerprint image with various challenging, a reliable pre-processing stage and an efficient feature extraction are needed. Segmentation is one of the most important pre-processing steps in fingerprint identification followed by image alignment, and enhancement. We improve a common enhancement technique based on STFT analysis by replacing the used segmentation technique which based on thresholding the energy map, with another one based on morphological operation. We use modern multiresolution techniques; Curvelet, Wave Atoms, Shearlet transforms in extracting distinctive features from the enhanced fingerprint images in a new methodology. The selected features are matched through multiple classifier techniques. We use the Minimum Distance Classifier, K-Nearest Neighbour, Self-Organizing Map and Support Vector Machine. We compare between all these classifiers with respect to the various feature extraction techniques. We test our methodology in 114 subjects selected from a very challenges database; CASIA-FingerprintV5; and we achieve a high recognition rate of about 99.5%.