Premium
Estimating age from photographs of children's faces
Author(s) -
DeLeon Valerie Burke,
Chollet Madeleine B,
Monteiro Stephen,
Chung Sarita
Publication year - 2011
Publication title -
the faseb journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.709
H-Index - 277
eISSN - 1530-6860
pISSN - 0892-6638
DOI - 10.1096/fasebj.25.1_supplement.866.3
Subject(s) - computer science , scale (ratio) , centroid , process (computing) , variable (mathematics) , artificial intelligence , test (biology) , face (sociological concept) , set (abstract data type) , statistics , geography , cartography , mathematics , mathematical analysis , paleontology , social science , sociology , biology , programming language , operating system
Reuniting children separated from their families following mass disasters is a primary goal for emergency management teams. We are developing a system to facilitate this process based on automated extraction of information from photos of children collected by emergency personnel. One of our goals has been to determine the best method for estimating age of a child from such a photo, obtained under adverse field conditions where standardized collection protocols cannot be assured. Our sample consisted of 110 children of known ages (2 mos – 16 yrs). Photos of the face were collected in frontal view. We manually collected a set of 17 landmarks using ImageJ. We developed a novel approach to establish scale in each photo based on our expectation that iris size does not vary with age. Right and left sides were averaged to reduce the effect of positioning error. Landmarks were superimposed, and we used multiple regression to test the predictive value of these facial shape variables for the estimation of age. Our results allowed us to predict age +/− 2.5 years (95% CI). Centroid size was the most important variable in predicting age, highlighting the importance of our novel approach to establishing scale. We plan to extend these finding to landmarks that can be automatically extracted from facial photos, allowing a fast and efficient method to limit the number of photos that families must review during the reunification process. Research was funded by HRSA EMSC Targeted Issue 5H34MC105750300.