QDFCS: A Comprehensive Database for Visual-Sketch Correspondence in Forensic Facial Identification

Abstract

Facial sketches have been used widely by law enforcement organizations to aid in identifying (and possibly capturing) individuals suspected in criminal activities. The sketches are identified as forensic sketches (drawn by hand by forensic artists, based on verbal descriptions from the victim and/or other eyewitnesses, or composite sketches (produced using computer software). Initially, such sketches were circulated on available media and in public places to enable viewers to identify the faces and hence report the identity of the subject. With the advent of closed-circuit cameras, the process of identification has become more dependent on algorithms designed to match faces between visual and sketch domains. This paper presents a database containing a set of sketch images (viewed, semiforensic, forensic, and composite) of the subjects whose visual images were previously reported in the QDF (Quality Dataset for Distance Faces) database. The present database is named as the Quality Dataset for Distance Faces and their Corresponding Sketch (QDFCS). The database has been evaluated using popular face recognition algorithms. The use of QDFCS in the training and evaluation of face recognition algorithms will be a significant contribution towards forensic facial identification.