Fingermark Image Quality Assessment with Random-Forest Classifier

Fingermarks, also known as latent fingerprints, which can be found on crime scenes, are used by law enforcement agencies to identify suspects. Dactyloscopic experts compare fingermarks from a crime scene with fingerprints of suspects taken under controlled conditions and stored in forensic databases. Only fingermark images of adequate quality can result in a conclusive match. Machine-learning techniques assessing the quality of fingermark images can support the tedious and time-consuming work of forensic experts. We propose a random forest model that classifies fingermark images based on handcrafted features into two classes indicating whether the images are of value for identification or not. This helps to ensure a sufficient quality of fingermark images to be examined.