Children Longitudinal Face Recognition Using Random Forest

Abstract

The children face recognition system plays a vital role in application towards the track and recognize the children who are missing at a young age due to child trafficking or kidnapping. In this study, an attempt is made to find the rate of recognition of a young child face images from age 1 to 15 years, where the frequency of facial growth is higher in this interval. To study the effectiveness of the problem, a dimensionality reduction technique is adopted such as PCA followed by random forest classifier as one method and deep convolutional neural network as another method. The proposed model was vindicated with a relatively large dataset that was created to address this problem. The total number of longitudinal face images of 47 children each 12-15 years was 685 and each year consists of a single sample. Experimentation was carried out by varying the number of decision trees and the number of classes for efficacious analysis of the problem. The procured results were promising with deep learning classification techniques with 93% of mini-batch accuracy.