Face Recognition of Children Using AI Classification Approaches

Abstract

According to the Census of India, children below 6 years of age have 13.12 % of the total population [21]. Security and healthcare of children is an important aspect of all countries. However, the span from newborn to toddler, recognition of them is still challenging using biometrics [1]. Recognition of children based on biometric can be used to minimize problems related to their security and health care. We require a well suited biometric based system for recognition of children. As per our knowledge, the accuracy of children identification in multiple sessions is not yet studied. Jain et. al studied Infants identification, achieved the accuracy of 38.44 % below 4 weeks and 73.98 % for above 4 weeks. In this paper we are presenting various face recognition techniques accuracy using different classification techniques for children. For this study the database of 90 subjects for the same session and 50 subjects of cross session is collected. Span between the data acquisition sessions is 2 months to 1 year. Detailed study of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) feature extraction using different machine learning classifiers on children database of the same session and cross session is carried out. Also, the Convolution Neural Network (CNN) optimized model with data augmentation is used to compare the machine learning and Deep learning classifier on our database. Same session accuracy of 98 % is achieved with LDA feature extraction and 5 -fold cross validation with LDA classifier. Cross session accuracy of 94 % is achieved with LDA feature extraction and Logistic regression classifier with 5- fold cross validation.