Back to Search Start Over

Wits Face Database: Reliability of forensic facial identification from CCTV recordings with implications for admissibility to court

Authors :
Bacci, Nicholas
Davimes, Joshua
Steyn, Maryna
Briers, Nanette
Publication Year :
2022
Publisher :
Open Science Framework, 2022.

Abstract

The human face is important in social, cultural and recognition contexts. Many research fields make use of faces to understand human interaction and identify individuals. Studies relying on facial image data often make use of ad hoc datasets specifically created for those studies as there is a dearth of large scale controlled and matching facial image databases. Actualistic (taken in a real life, natural setting) and standardised databases of facial images can be of extreme value to many research areas, such as facial identification and recognition. While multiple face databases are available, the majority, if not all, are developed in order to address very specific questions and hypotheses with limited standardisation, severely limiting their potential applicability. The Wits Face Database was developed as a generic, yet actualistic dataset of facial images obtained from consenting young adult South African male individuals. This database consists of high resolution standardised facial photographs (3264 x 4080 pixels) and corresponding closed-circuit television (CCTV) recordings of male South Africans under different camera conditions. A total of 6220 standardised (clothing and background controlled) and natural (visible clothing and out of focus background) facial photographs of 622 matching individuals in five different views are included. Corresponding CCTV footage of 334 of these individuals is also included. Across both the CCTV recordings and the photographs, the faces were captured in five different views: anterior, left 45-degree, left lateral, right 45-degree, and right lateral. The CCTV recordings were grouped under the following actualistic conditions: a standard internet protocol (IP) CCTV set-up, a low-resolution analogue CCTV set-up, an eye-level IP CCTV system, and the standard IP CCTV set-up with the addition of sunglasses and caps for target individuals. A detailed description of the composition and acquisition process of the database will be made available in a database descriptor publication format. The database is available strictly for non-commercial scientific research following approval of a formal application, assessed by the School of Anatomical Sciences’ Collections Committee within the University of the Witwatersrand. A total of 622 participants were photographed. Each participant had 10 photographs taken in different views and at two different distances (1.5m and 5m). The total number of unique photographs included in the database is 6220. All photographs were captured and stored both as native RAW and .jpeg formats. The effective duplication of photographs in these two formats results in 12440 images stored as part of the database. A total of 334 participants were also recorded using different CCTV cameras from the Wits Protection Services surveillance system. An approximate 30 seconds of footage was captured for each participant and extracted as .mp4 videos. The total approximate size of the database is 292 GB. Approximately 92 GB of the database consists of the CCTV footage with the remaining 200 GB being the photographs in all modes of capturing (5 views and two camera distances).

Details

Database :
OpenAIRE
Accession number :
edsair.doi...........0c6d405355af11e59a20bcad9346daa6
Full Text :
https://doi.org/10.17605/osf.io/msdbc