PAPER GENERAL; DIGITAL & MULTIMEDIA SCIENCES Kristin Norell, 1 Ph.D.; Klas Brorsson L€ ath en, 1 M.Sc.; Peter Bergstr€ om, 1 Ph.D.; Allyson Rice, 2 B.S., B.A.; Vaidehi Natu, 2 Ph.D.; and Alice O’Toole, 2 Ph.D. The Effect of Image Quality and Forensic Expertise in Facial Image Comparisons* ABSTRACT: Images of perpetrators in surveillance video footage are often used as evidence in court. In this study, identification accuracy was compared for forensic experts and untrained persons in facial image comparisons as well as the impact of image quality. Participants viewed thirty image pairs and were asked to rate the level of support garnered from their observations for concluding whether or not the two images showed the same person. Forensic experts reached their conclusions with significantly fewer errors than did untrained participants. They were also better than novices at determining when two high-quality images depicted the same person. Notably, lower image quality led to more careful conclusions by experts, but not for untrained participants. In summary, the untrained participants had more false negatives and false pos- itives than experts, which in the latter case could lead to a higher risk of an innocent person being convicted for an untrained witness. KEYWORDS: forensic science, information science, biometric identification, facial image comparison, CCTV, image quality Images of perpetrators in surveillance video footage are used as evidence in court both in Europe and in the United States. This is done commonly by showing the images to a jury (or similar) and letting these untrained individuals decide, based on their own judgment, whether or not the image depicts the sus- pect. It is also common to use the conclusions of a forensic expert as part of the evidence. Two important questions arise in both ways of handling the footage. These questions bear on the likelihood that the identification judgment will be correct. First, is there an accuracy difference between trained persons (forensic experts) and nontrained individuals (jury or similar) when mak- ing identity comparisons? Second, how does image quality affect the accuracy of trained versus nontrained individuals? Previous studies have examined these questions, but have not always arrived at converging conclusions. Wilkinson and Evans (1,2), for example, examined the accu- racy of forensic experts and nontrained persons on the task of matching identity between CCTV footage and facial image com- parisons (FIC). They asked two facial comparison experts and a group of 61 nontrained persons (students and university staff) to perform the task. Specifically, a video clip with one person was shown next to a pool of reference images. The video clips were of two kinds: one with the entire head shown, and the other with the person wearing a baseball cap to obscure the frontal hairline. The task of participants was to determine which, if any, of the subjects in the pool was the person seen in the CCTV footage. The results indicated that the experts performed better, both on measures of true acceptance and correct rejection. Experts also had fewer false acceptances and rejections for video clips in which the entire head could be seen. In other work, Lee et al. (3) examined the difference between trained and nontrained participants on very poor-quality CCTV footage. Their results indicated that the two groups of participants performed equally well on an identification task. Similar findings were obtained by Burton et al. (4), who asked experts and nonex- perts to identify unknown persons. In this case, a series of video clips was shown first. Next, each participant was asked to deter- mine from a pool of reference images which subjects had appeared in the video clips. This method is similar to the case in which an eyewitness must identify a person who he/she might have seen pre- viously. It is a different type of task, however, than the task usually given to a forensic expert, who must typically evaluate whether or not a person in a surveillance footage is the suspected person. Bruce et al. (5) showed that matching even a high-quality video image to a pool of reference images is a difficult task. The participants were correct only 70% of the time if they had no knowledge whether the person in the video image was present among the pool of images or not. Jenkins et al. (6) provided fur- ther evidence to conclude that matching face identity in images of unfamiliar people can be extremely challenging and error prone. They constructed an array of 40 images of two individu- als. The image array was presented to participants in a poster- type display, with the identities randomly intermixed, but simul- taneously visible. All images showed frontal views, but the pho- tographs varied in illumination and in the facial expressions showed. The age of subject varied also by approximately 10 years. Participants asked to determine the number of identities 1 Forensic Audio and Digital Imaging, Swedish National Laboratory of Forensic Science, SE-58194 Linkoping, Sweden. 2 US Department of Defense, School of Behavioral and Brain Sciences, University of Texas at Dallas, 800 W. Campbell Road, Richardson, Texas 75080-3021. *Presented in part at the University of Texas at Dallas, May 10, 2012, in Dallas, TX; and at the IEEE Biometric Summer School, June 2013, in Alg- hero, Italy. Received 29 Aug. 2013; and in revised form 12 Feb. 2014; accepted 28 Feb. 2014. 1 © 2014 American Academy of Forensic Sciences J Forensic Sci, 2014 doi: 10.1111/1556-4029.12660 Available online at: onlinelibrary.wiley.com