Children Facial Growth Pattern Analysis Using Deep Convolutional Neural Networks

In this work, a study of facial growth rate analysis from age 1 to 12 years on children’s longitudinal face dataset is presented. Our dataset created during this study is used to experiment with the proposed work. It is made up of 384 longitudinal face photographs of 32 children, each of whom has exactly 12 years of face images starting at the age of 1, with a single sample every year. The features are extracted from the top 10 pre-trained Convolutional Neural Networks. The distance between two successive ages of a child has been computed using Euclidean Distance. The distance measures for 32 children are computed, and the average of all 32 children concerning each patch is taken. Seven different patches of face images have been utilized for analysis. For better visualization of the obtained results, a graph for an average of 32 children distance measures from 1 to 12 years of two consecutive ages of a child is plotted. For brevity, the growth rate pattern in children is flattened smoothly from age 3 to 9. The author can declare that above age 9, face biometric can be adopted for efficient face recognition from this extensive experimentation.