Selecting discriminative CLBP patterns for age estimation

Face analysis is of great interest in the context of digital signage to understand soft biometric of a person. Among the others information gathered from a face, the age of a person is still an open challenging problem. Face representation takes an important role for real time age discrimination. LBP descriptor and the related variants (e.g., CLBP) have been demonstrated to obtain the state-of-the-art performances in this field. In this paper, building on the CLBP representation, we propose a method to select the most discriminative CLBP patterns to represent faces for age classification. Experiments confirm that the proposed method improves the age classification accuracy by reducing computational costs both in terms of space and time.