This paper describes another algorithmic procedure for classifying fingerprint patterns. The authors are interested in automating only the first level of classification; that is, dividing input prints into their major structure, such as loop, whorl, or arch.
Even though some parts of the paper exhibit awkward English syntax, the paper describes the methods used with both succinctness and clarity. The image of the print is binarized and coded by the major directions present in subregions. The coded directions are then run through a relaxation process to remove noise. Singular points (points where different directional flows meet) are then detected, based on a previously published criterion. Finally, starting at each singular point and tacing over flow lines, heuristic rules are applied to classify each print into one of the structural classes.
It is hard to evaluate the final results of a paper such as this. The method was run on 94 sample fingerprints and achieved a probability of error of approximately eight percent. However, the method may have been tailored for this set of samles. No comparison can be made to the numerous competing methods published in the literature, either in terms of classification accuracy or computational complexity.