Advanced feature extraction algorithms for automatic fingerprint recognition systems
MetadataShow full item record
Fingerprint identification is commonly used in authentication and forensic applications and is by far the most common biometric modality in use today. We have developed several novel image processing routines to address the problems with current Automatic Fingerprint Identification Systems (AFIS) which fail to perform when the fingerprint images are of poor quality. Current image enhancement methods are not sensitive to the type and quality of fingerprints. We have developed an adaptive filter, which automatically selects the preprocessing parameters based on the image quality. Our method uses a band-limited ring-wedge spectrum energy of the Fourier transform to characterize the global flow pattern of the ridges. It combines features from both the frequency domain and the spatial domain to provide a quantitative estimate of the quality which is then used to selectively enhance fingerprint images. Fingerprint ridge flow patterns are categorized into two types: pseudo-parallel ridges and high-curvature ridges surrounding singular points for this purpose. The enhancement filter uses a coherence map to locate the high-curvature regions (delta and core) and Gaussian kernels whose sizes adapt to the local ridge topology. Segmentation of the regions of interest is another important preprocessing task. Blockwise segmentation methods described in the literature cannot remove spurious minutiae (features) present at the boundary of blocks. We have developed a novel algorithm that uses the strength property of the Harris corner point which is usually higher in the in ridge regions. These points are filtered based on thresholding (Otsu's method) of the response of the Gabor filter to identify the outliers. The foreground boundary contour is located by a convex hull of the remaining Harris points and spurious boundary minutiae are discarded. Thinning based minutiae detection algorithms are computationally intensive and have the problem of minutiae shifting (by the ridge width). Irregularities and aberrations in the ridge flow also tend to cause spurious minutiae. We have developed two thinning-free minutiae extraction methods based on chain-code contour tracing and run length scanning. Our image processing framework has been tested on the FVC2002 DB1 and DB4 datasets each containing 800 fingerprints and compared with benchmark software developed by the National Institute of Standards and Technology (NIST). We have achieved an equal error rate (EER) improvement of 15% on the DB1 set and 37% on the DB4 set in the comparative tests.