Detection of A Stable Age in Children for Face Recognition Application

Abstract

In this work, a novel approach for finding an accurate age of a child to be adapted for an automatic face recognition application is proposed. Our proposed approach initially uses Viola–Jones method to automatically locate a key point on facial components namely nose, eyes and mouth. We extract the intensity value in the region across these key points, find the row mean intensity value in the region of each component and sorted. Then, the corresponding index value of the sorted elements are preserved and represented as a Feature Vector (FV). By using the FV of each component, we find the correlation coefficient between two consecutive ages of a child. Then, the First Order Difference (FOD) of correlation coefficient of two consecutive ages is computed. The Cumulative Histogram (CH) of FOD is estimated and average CH of all components is determined. If the difference between average CH of two consecutive ages is negligibly small, then we considered that age as stable age for face recognition application. The proposed model has been validated on a reasonably sized dataset, which is created during this work. It consists of totally 421 longitudinal face images of 32 children each having 12–15 years face images, single sample per year. Kappa coefficient is used to measure the goodness of the model and results with 50% accuracy.