Fast Fourier transform for fingerprint enhancement and features extraction with adaptive block size using orientation fields

Extracting minutiae from a digital fingerprint is a crucial step in a fingerprint-based recognition systems. This work deals with poor-quality fingerprint images containing broken ridges. The enhancement stage connects broken ridges and is essential for extracting correct minutiae. We use a FFT variant [8] for this stage, but, to truly benefit from FFT in a block, it is essential to determine a suitable block size, depending on ridges orientation field. We propose to use a quadtree to partition the ridges orientation field into homogeneous blocks. A block is homogeneous when at least seventy percent of its ridges orientations are within ten degrees. Another issue addressed in this article is the choice of a suitable neighborhood window size W for computing orientation field image, depending on the fingerprint image quality. The performance improvements of our algorithm are evaluated and compared with standard measures MSE, PSNR and GI, on databases DB1 to DB4 of FVC2004 and NIST special database SD302d