CCTV quality assessment for forensics facial recognition analysis

Abstract

Closed-circuit television (CCTV) is used to perform surveillance recordings, and it is one of the most common digital devices that provide digital evidence for the purpose of forensic analysis. In video forensic analysis, the footage with the target subject or object is extracted out from the CCTV recordings for further analysis. However, the quality of these recordings are often poor due to several factors, such as the type of the camera, the configuration, and also the position of the camera. The results of forensic face recognition depend highly on the quality of the CCTV recordings. Poor quality of CCTV recordings would reduce the confidence level of the face recognition result, thus would not make a strong evidence to be presented in a court of law. The objective of this research is to conceptualise a framework for quality assessment in CCTV evidence to be used in forensic face recognition analysis. The method of this research was divided into two phases. Initial phase covered CCTV evidence testing phase where the experiment was done based on different types of CCTV camera with different resolutions, and distances between the subject and the camera. In the second phase, the face of the subjects were compared to the face taken during the enrolment phase. The score obtained from the forensic face recognition system would be based on the camera resolutions, types of camera, distances, and also the changes of ranking score after applying the enhancement process such as Bicubic to the facial images. The results were analyzed for quality assessment towards these parameters. In general, the evaluation of scoring and ranking decreased as the distance increased. The face also could not be detected by the system when they were taken more than 5 meters distance from the camera. The highest score of 5.95 was obtained by using resolution 1280 × 720 at distance of 3 meters taken by camera model ACTI E62. The Bicubic enhancement method improved the scoring and ranking especially with the camera model that have low resolution modes.