Testing the Matching Capabilities of Megvii's Face++ Using Age-progressed and Real-life Images




Brown, Emily

Journal Title

Journal ISSN

Volume Title



The National Center for Missing & Exploited Children (NCMEC) assisted law enforcement with over 29,000 missing children cases in 2019 and has completed more than 6,800 age-progressed images in its history of working on long-term missing children cases. There is currently little research on the topic of age progressions and their impact on facial recognition algorithms specifically when comparing real-life images and digitally produced age-progressed images of the same individuals. The goal of this study was to determine if a facial recognition algorithm could accurately match and generate a missing child's age-progressed image in a list of top 5 candidates when using the child's real-life image as the probe image for the search. Another goal of this research was to determine if there were any differences in the likelihood of matching based on the age of the missing child and the age variation between the child's real-life image and his or her respective age-progressed images. The age-progressed and real-life images of 347 children who went missing between the ages of 1 to 20 were included in the study. A gallery of images (called a FaceSet) was created and included the age-progressed images of all 347 missing children. The missing children's real-life images were searched against the FaceSet using Face++'s Search API and the top 5 matches for each person were generated. Every child was categorized as being in the 'older' group (>=13 - 20 years old) or 'younger' group (<13 years old) based on the age the child was when he or she went missing. The results of the study showed that the confidence scores of matches are higher for older children and there is a greater likelihood of matching for older children. The results of the study also demonstrated that the age-progressed images closest in age to the age of the missing child have a greater chance of being matched as compared to the age-progressed images with more age variation.



Age progression, Facial recognition


Brown, Emily. " Testing the Matching Capabilities of Megvii's Face++ Using Age-progressed and Real-life Images." George Mason University. Fairfax, VA. 2020.