Date of Graduation


Document Type


Degree Type



Eberly College of Arts and Sciences


Forensic and Investigative Science

Committee Chair

Keith Morris

Committee Co-Chair

Suzanne Bell

Committee Member

Afzel Noore


One of the most common types of evidence recovered from a crime scene are latent fingerprints, however these impressions are often of low quality. The quality of a latent fingerprint is described as the degree to which the ridge details can be observed. If the quality of the latent fingerprint is very clear, a minutiae-based matching algorithm with automatic extraction may detect and utilize the minutiae that are truly present in the fingerprint. However, if the impression is of poor quality, the minutiae-based matching algorithm's automatic extraction may detect fewer features and could completely miss features resulting in the return of an unrelated candidate. The aim of this research was to determine a method to improve the match score of latent fingerprints by removing the bad quality regions, where both a subjective and objective methods were utilized. The subjective method utilized the predetermined quality categories of "good," "bad" or "ugly" to assign a latent fingerprint. After classification, each impression was processed by AdobeRTM PhotoshopRTM and four quality areas were serially removed. In the objective method, each latent fingerprint was assessed with NFIQ algorithm and then MINDTCT algorithm. The MINDTCT algorithm provided a quality map that was used to remove successive portions of each latent fingerprint. The resulting new images from both methods were compared to a database using the two different minutiae-based matching algorithms: AFIX TrackerRTM and BOZORTH3.;The results were examined utilizing the statistical methods of receiver operator characteristic (ROC) curves, area under the ROC curve (AUC), cumulative match characteristic (CMC) curve, Wilcoxon signed-rank test, Spearman's rank correlation and the comparison of the removal methods. ROC curves and the resulting AUC were able to determine that the AFIX TrackerRTM program is a reliable performer with high AUC values, while the BOZORTH3 minutiae-based algorithm did not perform well with low AUC scores of around 0.5. The results produced from the CMC curves showed that the subjective method produced higher rank 1 and top 10 rank identification than the objective method, contrary to what was hypothesized. The correlation scores showed the manual and automatic extraction were weakly correlated to one another. However, a very weak to no correlation between the algorithms of the BOZORTH3 and AFIX Tracker R was observed. The comparison between the subjective and objective methods of removal showed the examiner allowed for a more conservative removal of the fingerprint than the objective method. With this result in connection with the CMC curve results shows that being more conservative produces higher rank 1 and top ten rank identification scores.