Investigating the effect of dimensionality reduction in a sign language recognition task

Details

Supervisors

Saerens, Marco

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

In today’s world, communication is one of the most important aspects of our society. With the rise of new technologies, communication between people has become even eas ier. However, for a certain portion of the population, it is not the case. Indeed, the deaf community still encounters great difficulty communicating with other people, as few people speak sign language. Our master thesis’s objective is to study the effect of dimensionality reductions on machine learning models for isolated sign language recog nition. Three dimensionality reduction techniques were investigated: principal compo nent analysis, auto-encoder, and singular value decomposition. Then, for each of these techniques, we explore the performance of several models, such as support vector ma chine, artificial neural network, recurrent neural network, gated recurrent unit, and long short-term memory, to see which model works best with which dimensionality reduction technique. Lastly, we investigate the transformation of our data into images in order to apply convolutional neural network.