Automatic recognition of facial expressions

Abstract

Image classification refers to the task of assigning an input image one label from a fixed set of categories. This is one of the core problems in Computer Vision despite its simplicity, has a reputation of being extremely difficult to implement due to the lack of labelled datasets. However, it has a large variety of practical applications especially in the field of facial expression recognition. The ability to recognize facial expression recognition automatically enables novel applications in human-computer interaction and other fields, therefore this leads to intensive and active research by many to create a competent and accurate image classfier through the utilization of Deep Learning techniques and formation of an ensemble of deep Convolutional Neural Networks (CNN). In this project, the focus would be on the implementation of a simple semi-super supervised model which would be an extension of the customised supervised image classifier model while making effective use of the labelled and unlabelled data. Next, it also aims at providing insights on the transferability of deep CNN features to address the long-standing problem of unlabelled images in huge datasets. Lastly, the paper would also study the impact of hyperparameter tuning on the model training process and image classification results.