Normalized cross-correlation based global distortion correction in fingerprint image matching

We present a pre-processing step for minutiae based fingerprint verification to perform distortion correction. We first detect the core-point of the enrollment image, select a Region of Interest (ROI) around it and find its corresponding region in the test image using normalized cross-correlation method. Then using core-point and the lower-right corner point of ROI, we build fixed and moving point sets to estimate the transformation parameters. Then we perform global distortion correction using nonreflective similarity transformation on test image. Comparing to traditional approaches which first extract minutiae points then perform alignment, we impose global tuning of the scale, translation and rotation variances even before minutiae extraction. Also, instead of extracting all minutiae points on entire image, we only consider the common intersection region of enrollment and transformed test images. Thus, we reduce the complexity of overall minutiae extraction task. Using widely used FVC2002 dataset, we compare our method with the result of [4] and one commercial fingerprint verification SDK [15]. The experimental results show us our method performs better. The equal error rate (EER) of our algorithm on FVC2002 is 6.0%, 6.0% and 7.0% for DB1, DB2 and DB4, while 14.0% for the most challenging DB3.