Age, Gender Prediction and Emotion recognition using Convolutional Neural Network

Abstract

Automated age and gender detection has been generally used in our daily lives that we come across, majorly in a person to computer interaction, visual surveillance, biometric analysis, electronics and other applications of commercial use. By recognizing the emotions of a person, we can improve the recommendation system. The existing methods have quite satisfying performance on real-world images if facial expressions of input image is neutral or calm, it lacks significantly in age prediction when facial expressions are altered. The dataset was obtained from IMDb-WIKI for age-gender classification and Fer2013 dataset for emotion recognition from kaggle. This project has two models, one for age-gender prediction using wide resnet architecture and the other model is trained for emotion recognition using conventional CNN architecture. An improvement in the performance of these tasks was observed by using the convolutional neural-nets (CNN).The accuracy of the wide-resnet model is 96.26% and for the emotion recognition model, accuracy achieved is 69.2%.