Comparison between face recognition algorithms in acessibility context

Author(s)

Douglas Eduardo Parra

• Document type: Master's Dissertation
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Instituto de Computação
• Defense date: 2014-06-27

Examining board members

Siome Klein Goldenstein; Jacques Wainer; Eduardo Valle

• Advisor: Siome Klein Goldenstein

Abstract

In this master dissertation is shown a comparison between three face recognition algorithms within the context of accessibility to the Microsoft project in partnership with FAPESP for the people recognition module using Microsoft Kinect and sensory substitution. The k-Nearest Neighbours algorithm, with Histogram of Oriented Gradients, was employed as a basis for being a simple and low computational cost. The Eigenfaces and Local Binary Patter Histogram algorithms were compared with the previous one in four experiments. Initially, the Project Vision for the Blind and its different modules is described. This project was developed by a team in Brazil which achieved good results for navigation and face recognition modules, always with the idea of using audio 3D to convey the desired information to the user. It will next be shown a review of the state of art with projects within the context of accessibility e sensory substitution, pointing out its limitations. Immediately after it is done a review about the three face recognition algorithms used and then how the image database from this project was created. Good results were achieved with the three algorithms although there are significant differences among them. Eigenfaces and Local Binary Pattern Histogram, for being more complex techniques than k-Nearest Neighbours, reached recognition rates with half of the resources than the last one use to get close to the result values, with Eigenfaces being the fastest. Nonetheless, for being a simple technique, it is worth take note how good k-NN executes the same task and could be used in the project module (AU)

• FAPESP's process: 12/22653-3 - Vision for the blind: translating 3D visual concepts into 3D auditory clues
• Grantee: Douglas Eduardo Parra
• Support Opportunities: Scholarships in Brazil - Master