Vision-based gender and age classification

Abstract

Recently, artificial intelligence (AI) has become closer to human life. For example, an intelligent home helps us manage our home appliances and makes our lives more convenient. One of the usages of AI is doing object classification, classifying gender, and evaluating age. Once we estimate gender and age, we can do more in-depth research in multiple fields, such as micromarketing. In the past research, most of them use headshot photos for training convolutional neural networks (CNNs), which need to capture in front of the human face. However, it is not easy to capture headshot photos for evaluation in the wild. Therefore, I am using the gait energy image (GEI), created by human gait, to train and test my CNN model. This project is using two datasets, OULP-Age and OU-MVLP, which OU-ISIR Biometric Database provides. In this project, I will combine and optimize the programs and models from other students. Also, using a new dataset, OU-MVLP, I will create a new model. For the result, I have combined and changed the structure of programs from the first two students. Based on OU-MVLP, I have succeeded in training a 3-layer model. Our model can achieve lower mean absolute error in age classification than three state-of-the-art methods.