Personal identification through the use of fingerprints is probably the most used biometric modality, and its accuracy is crucial in applications critical to society and business. Improved methods for preprocessing, feature extraction, and matching have all led to the high accuracy of today’s fingerprint recognition systems. The authors here propose a method for enhancing fingerprint images to improve matching accuracy.
They begin by briefly discussing the extraction of minutia points from a fingerprint image; these features are locations where the fingerprint ridges either end or split into two, along with the angle of the ridge at the end or split point. While not all fingerprint recognition is done on the basis of minutia points, or on minutia matching alone, it is the most common feature representation, especially in the forensic (law enforcement) field. Most enhancement operations are conducted either in the spatial or transform domain. This paper properly points out that much research has been published on methods for fingerprint image enhancement, and the references mention a sampling of the work in this field.
The proposed method consists of a spatial enhancement operation followed by a filtering in the transform (spatial frequency) domain. The spatial enhancement proposed is a form of localized image normalization, operating on small neighborhoods within 8-by-8 blocks of pixels. Based on the image database used for testing, it’s likely that this value is optimized for fingerprint images captured at 500 dots per inch (dpi); newer systems that use 1,000 dpi may require larger window sizes. Once the image is locally normalized, radial and angular filters are applied to enhance spatial activity within a range of spatial frequencies. This is an alternative to the frequently used selective Gabor filter. The output image shows generally good contrast and good preservation of the key features in the image. The authors test their method by performing extraction and matching on the enhanced images. The results in equal error rate (EER) are positive, though it should be noted that the improvement attributable to the enhancement may depend strongly on the extraction and matching methods used.
The paper is a complete and understandable description of an enhancement method for fingerprint images. As another addition to a well-researched field, it is more contributory than groundbreaking. Systems that rely on level 3 fingerprint detail (pores), that use 1,000 dpi resolution, or that do not use minutia for matching may not find the method useful, but many applications could possibly benefit from the proposed algorithm.