A novel feature extraction algorithm from fingerprint image in wavelet domain

Abstract

The robustness of a fingerprint authentication system depends on the quality of the features extracted from the fingerprint image. For extracting good quality features, the quality of the image is to be improved through denoising and enhancement. In this paper, a set of invariant moment features are extracted from the approximation coefficient in the wavelet domain. Initially the fingerprint image is denoised using Stationary Wavelet Transform (SWT), a threshold based on Golden Ratio and weighted median. Then the denoised image is enhanced using Short Time Fourier Transform (STFT). A unique core point is then detected from the enhanced image by using complex filters to determine a Region of Interest (ROI), which is centered at the enhanced image. Then the ROI is decomposed using SWT at level one of Daubechies wavelet filter for extracting efficient features. The decomposed image is partitioned into four sub-images to reduce the effects of noise and nonlinear distortions. Finally a total of four sets of seven invariant moment features are extracted from four partitioned sub-images of an ROI of the approximation coefficient as it will contain low frequency components. To measure the similarity between feature vectors of an input fingerprint with the template stored in the database, the Euclidean Distance is employed for FVC2002 dataset. Using a simpler distance measure can substantially reduce the computational complexity of the system.