Age group estimation on single face image using blocking ULBP and SVM

Since age implies essential individual information for human beings, age estimation has more and more applications in intelligent human-computer interactions and personalized recommendation in SNS, etc. However, precise age estimation based on single image is difficult due to diverse appearances among people, and irregular quality of sample acquisition. Based on general knowledge that wrinkles increase with age, Uniform Local Binary Patterns (ULBP) is always an effective texture descriptor, but it loses relative location information. In this paper, an age group estimation algorithm is proposed, where after efficient preprocessing, blocking ULBP is used to gain facial textures and a trained multi-class SVM is applied to fulfill age classification. The ages of subjects are divided into five groups: children (0-6), juveniles (7-18), youth (18-40), middle-aged (40-65), and old people (�?6). Experiments are implemented on FG-NET and Morph Aging Database and the estimation accuracy achieves 81.27%. © Springer-Verlag Berlin Heidelberg 2015.