Age estimation upon face image based on local binary patterns and a ranking approach

A new age classification algorithm is suggested, which is a modification of method developed by Chang et al. The algorithm is based on training of a set of binary classifiers. Each classifier estimates whether the person is older than a specified age or not. The age then can be simply calculated as a sum of outputs of all binary classifiers. Using local binary patterns as classification features, age prediction accuracy improvement is achieved, though classifier size is increased. A number of modifications, which decrease a classifier size and increase classification speed, but keep age estimation accuracy high, are proposed. Experiments on MORPH database showed mean absolute error from 4.52 to 5 years and classification time between 0.32 and 3.21 s, depending on parameters.