Fingermark quality assessment with deep learning ensembles
DOI:
https://doi.org/10.31449/upinf.182Keywords:
deep learning, fingermarks, forensics, quality assessmentAbstract
Quality assessment is an important step when trying to identify fingermarks from a crime scene. Often done in the scope of forensic investigation, it is performed by trained examiners and tends to be rather subjective. The goal of our work is to develop an automated fingermark quality assessment method, which would assist the examiners in their work. In this paper, we introduce modern deep learning techniques into the field of fingermark quality assessment, we evaluate the advantages and disadvantages of this methodology, and identify key aspects for further development in the field. We propose a new quality metric, which works by fusing individual predictions of an ensemble of deep models. The proposed approach provides improved prediction performance while reducing processing time by at least a factor of 15 compared to existing solutions.