|
V10 N1 Paper 1
|
Annals of the MS in Computer Science and Information Systems at
UNC Wilmington
|
Spring 2016
|
Child Face Recognition: Establishing Baseline Performance Metrics
Shivani Bhardwaj
Committee
Abstract
Automatic face recognition is a challenging task that has made significant advancements over the last decade against the problems of pose, illumination, and expression (PIE). There has also been an improvement in the systems against the challenges of time displacement, also known as aging. Aging results in face variation, which affects the performance of a face recognition system. However, this work uncovers a problem of aging that has not, as of yet, received attention from the research community. This body of work explores the challenges of face recognition, and by inference soft biometrics or facial analytics, for sub-adults, the population of faces that exists between ages 0 and 18 years. The velocity of craniofacial morphology in the sub-adult population can be quite aggressive as compared to the changes in the adult population, and such rapid changes in both the hard (bony) tissue and the soft tissue can cause demonstrable degradation in face recognition as well as facial analytics. The objective of this work is to highlight the challenges with the issues of longitudinal face recognition and the soft biometrics on sub-adults. Further, this work establishes the difficulty of this problem by establishing a seminal baseline as well as the critical biological underpinnings for the cause of the problem of child face recognition. This work institutes the baseline against the largest longitudinal sub-adult corpus created to date on a set of commercial matchers. The impact of variation caused because of aging on face recognition technology was explored with commercial face recognition systems, Cognitec’s FaceVacs v8.50, Rank One Computing v1.20, and Neurotechnology’s Verilook v6.0. The performance of these commercial systems reveals the challenges associated with this problem: average performance across the systems tested for identification of a person from its younger image: 25.06% (Rank-1) and 47.03% (Rank-20) with a dataset of 501 subjects.
download
(pdf)
Recommended Citation:
Bhardwaj, S., Ricanek, K, Cummings, J., Pence, T. (2016) Child Face Recognition: Establishing Baseline Performance Metrics. Annals of the Master of Science in Computer Science and Information Systems at UNC Wilmington, 10(1) paper 1. http://csbapp.uncw.edu/data/mscsis/full.aspx.
|
V10 N1 Paper 1
|
Annals of the MS in Computer Science and Information Systems at
UNC Wilmington
|
Spring 2016
|
