ABSTRACT The natural process of aging affects human faces in a variety of ways. These effects make many automated processes, as well as human-driven processes, difficult. The need to recognize faces after a gap of several years, though, is likely to be the norm in many law-enforcement and other face-recognition application environments, making consideration of aging very important. To date, though, scant research in modeling and face recognition has even considered the effects of aging. This work considers an active-appearance-model approach to “facial aging” of images of adults. Some similar work has been conducted concerning growth and development but not adult aging, which is a distinct and separate process. A sample from a new database, containing some of the largest age spans of any publicly available face database, is used along with active- appearance models to artificially age images of adult faces. A brief anthropological perspective of aging is presented, the application and results of active-appearance models to facial-image aging are given, and how these results could affect face-recognition and forensic applications is discussed. KEYWORDS Aging, face recognition, biometrics, and image synthesis. 1. Introduction Many aspects affect the appearance of a person’s face during the process of growing older. Although these changes in appearance arise from a variety of contributing factors that vary by elements such as lifestyle, race, and geographical region, they are often collectively termed “aging.” The associated changes may be subtle or large, depending on the individual and time span covered, but the changes can frustrate human and computer-based recognition of individuals where some span of time has passed between sample images. There has been little study of this to date, but devising effective models for including effects of aging in modeling and face recognition applications could yield significant improvements in a variety of areas. This paper discusses work that progresses toward automatic generation of accurate images of artificially "aged” individuals and also progresses toward improved face-recognition algorithms through the incorporation of face-aging models and techniques. 1.1. Effects of Facial Aging There are at least two areas that effective models of face aging could directly aid: automated computer face recognition and human use of hypothetically updated images in law enforcement applications. Currently, face recognition technology is heavily influenced by any change in a test image versus training images; this change could include aspects of pose, illumination, facial hair, and certainly aging [1, 2]. In addition, most face recognition databases in use only span a relatively short time period between sample images and thus cannot be used to study nor effectively demonstrate robustness of techniques to age variation [3, 4]. The capability to model accurate image and face model changes due to age could be used to update a face-recognition training gallery for improved recognition or even be incorporated directly into the face recognition algorithm. Knowledge of what changes occur and what features stay the same could also be used to focus on recognition techniques that produce age-invariant results. In addition to improving face recognition software, the capability to produce an accurately “aged” image of an individual could aid several fields. Currently such images are produced for forensic and other applications as well as for study of history and anthropology but are not produced through automated means or with particular accuracy and rigor. The primary technique used to date is creation of a drawing by a trained forensic sketch artist, incorporating some scientific data in a largely artistic approach to generating a facial image. Automated computer methods for generating face images are becoming more popular, and commercial software packages have recently become available, but these are primarily based on an artistic approach as well, rather than including detailed, specific models of accurate aging that could be applied to a variety of individuals [5, 6, 7]. The capability to produce images that accurately represent the aging-related changes that would have likely occurred to a particular individual since AUTOMATIC REPRESENTATION OF ADULT AGING IN FACIAL IMAGES E. Patterson, K. Ricanek, M. Albert, and E. Boone University of North Carolina Wilmington 601 S. College Rd., Wilmington, NC U.S.A. pattersone@uncw.edu, ricanekk@uncw.edu, albertm@uncw.edu, and boonee@uncw.edu