New M.E.Thesis Submitted from CSE

A NOVEL APPROACH FOR FINGERPRINT ENHANCEMENT USING SPATIAL AND FREQUENCY DOMAIN FILTERING by Mehak Sood 

Abstract

Fingerprints are one of the most popular biometric which find use in a wide number of applications. The fingerprint images gathered are generally of low-quality due to varying input contexts such as different pressure applied, dryness or wetness of hands, scars and so on. The low quality of these images affects the performance of the automated fingerprint identification systems (AFIS). Generally, the algorithms are developed using techniques from either the spatial domain or the frequency domain to enhance these low-quality fingerprint images. This research work proposes a novel algorithm to enhance the low-quality fingerprint using techniques from both the spatial and frequency domain. Firstly, coherence filtering using optimized rotation invariant anisotropic diffusion filter is done. This processing joins the broken ridges and removes the noise. Then, ridge filtering which uses a bank of Gabor filter in the spatial domain selected specifically according to estimated ridge orientation and frequency is done. This step helps in separating the ridges and valleys which improves the distinction between them and recovers the damaged regions but at the same time produces a blurring effect. Finally, frequency domain filtering using a bandpass filter which uses the re-estimated ridge orientation and frequency estimations from the spatial domain enhanced image is done to get the final enhanced image. This step removes the blurring effect caused by the spatial domain processing and helps in clearer distinction between the earlier merged ridge and valley regions. The proposed scheme joins the broken ridges, helps in clear distinction between ridges and valleys, recovers the damaged regions, removes the noise, improves the contrast, and smoothes the ridges.  To compare the performance of the proposed algorithm, some state-of-the-art techniques: coherence filtering, Gabor filtering, Fourier analysis, and a hybrid approach were implemented. The visual analysis shows that the proposed algorithm gives the best results as compared to some of the state-of-the-art techniques over randomly chosen images from public FVC2004 database.  The texture descriptor analysis shows that the proposed algorithm, improves the texture of the fingerprint image better than the other state-of-the-art techniques. Two feature extraction techniques namely the thinning technique and the mindct technique are used to compute the minutiae ratios. The minutiae ratio analysis is done separately for both the feature extraction techniques. The minutiae ratio analysis shows that the proposed algorithm achieves high True Minutiae Ratio (TMR) of 94.82% and low False Minutiae Ratio (FMR), Dropped Minutiae Ratio (DMR), Exchanged Minutiae Ratio (EMR) with thinning technique of feature extraction. Also, the proposed algorithm achieves high TMR of 90.45% and low FMR, DMR, EMR with the mindct technique of feature extraction. Moreover, the proposed algorithm outperforms the above stated stateof-the-art techniques for both the feature extraction techniques when comparison is done on the four sub-databases of FVC2004 separately using both the feature extraction techniques. The enhanced image will thus help in improving the performance of the AFIS as well.