Frequency Domain-Based Identification of Salient Regions for Minutiae Extraction in Fingerprint Images

This paper proposes a novel method for identifying all types of minutiae points on a clearly scanned fingerprint image using a unified strategy in the frequency domain. The core of our approach lies in the characteristic representation of fingerprint images in the frequency domain, where the band-pass filter region, devoid of minutiae points, serves as a baseline. Any deviations from this baseline, indicative of minutiae points, are detected as anomalies. To facilitate this, we develop a unique, fingerprint-specific, two-dimensional band-pass filter derived from the frequency domain representation of the image. Experimental results demonstrate that our method achieves a higher correlation with traditional minutiae detection strategies, underscoring its effectiveness. The proposed algorithm first removes noise from the fingerprint image using multi-resolution analysis, followed by a transformation to the frequency domain using fast Fourier transform. A composite band-pass filter is then designed based on the frequency domain representation, which is subsequently used to filter the image and extract minutiae points through morphological operations. Our findings indicate that the majority of minutiae points lie within salient regions identified by our approach, with an error rate of less than 2%.