Face classification using deep learning techniques

OData support
Supervisor:
Dr. Kővári Bence András
Department of Automation and Applied Informatics

The reason why I have chosen the classification of faces as the subject of my thesis is that the expectations from the smart electrical devices are becoming higher and higher, people want the devices to adapt to the current user much better. If a device could classify the faces according the emotions of it, it would result in a much more personalized behaviour.

While people are able to detect emotions on a face easily, it is not a trivial task for a computer. People can do this on visceral way, but the computers cannot solve this problem programmatically, so we have to seek for other techniques.

First of all I had to gain theoretical knowledge in neural networks for solving this problem, and afterwards by summarizing it the Reader can also have a deeper look into this kind of behaviour. I start with the most frequent basic concepts, like perceptron, multi-layer perceptron and activation functions, and then I introduce more complex structures as well, for example the convolutional neural networks and the corresponding types of layers. After that I use this knowledge to reach my goal, I create a model that tries to detect emotions by portraits. The accuracy of the model that I created is able to approach the human accuracy with the same dataset.

This given knowledge can help not only with the problem that I have described, but it also shows an approach to the Reader, that (s)he can use to solve similar problems in an easier way.

Downloads

Please sign in to download the files of this thesis.