http://ieeexplore.ieee.org/iel5/7192/19378/00895395.pdf

One of the main difficulties with using fingerprints (as evidence, record-keeping, etc) is that there are many factors that make sampling someone’s fingerprint inherently inaccurate (e.g. low quality scanning technology, issues with skin itself, etc). Thus a good fingerprint image-enhancing algorithm is something that would be very useful in this day and age. Until recently, such algorithms were inhibited by a very popular computational dilemma: efficiency or accuracy? The above paper offers an algorithm that boasts impressive results for both.

The basic idea is to use least squares to solve a few isolated, representative examples of the problem. To solve these representative problems, we need the help of an “expert” who can offer some tuning data given an image. The best filter for a given input image is found by minimizing the differences between the given pixels and the expert’s analysis (i.e. a least-squares minimization). After solving these isolated problems (albeit with the help of a human), we generalize the findings to apply to any given fingerprint image. Analyzing the image’s properties will help us discover which sub-solutions to use and how to use them.