Automatic latent value determination

Latent fingerprints are the most frequently encountered and reliable crime scene evidence used in forensics investigations. Automatic methods for quantitative assessment of a latent in terms of (i) value for individualization (VID), (ii) value for exclusion only (VEO), and (iii) no value (NV), are needed to minimize the workload of latent examiners so that they can pay more attention to challenging prints (VID and NV latents). Current value determination is either made by examiners or predicted given manually annotated features. Because both of these approaches depend on human markup, they are subjective and time consuming. We propose a fully automatic method for latent value determination based on the number, reliability, and compactness of the minutiae, ridge quality, ridge flow, and the number of core and delta points. Given the small number of latents with VEO and NV labels in two latent databases available to us (NIST SD27 and WVU), only a two-class value determination is considered, namely VID and VID̅, where the VID̅ class contains VEO and NV latents. Experimental results show that the value determination by the proposed method (i) obviates the need for examiner markup while maintaining the accuracy of value determination and (ii) can predict the AFIS performance better than examiners.