Investigation and enhancement of viola jones face detection algorithm performance using contrast-limited adaptive histogram equalization

• Main Author: Mulyadi, Erina
• Other Authors: Ma Kai Kuang
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/64737
• Institution: Nanyang Technological University

Face detection determines the presence and location of a human face in an image. Face detection is the first and the most important steps to the entire face analysis algorithms, including face alignment, face relighting, face modeling, face recognition, head pose tracking, face verification/authentication, facial expression tracking/recognition, gender or age recognition, and many more. [1] While this sound like a simple task for humans, it is still an extremely challenging task for computers. In the past few decades, face detection has been one of the top research topics. In 2004 Paul Viola and Michael J. Jones publish an article with a title “Robust Real-Time Face Detection”. The algorithm has been so successful that today it is very close to being the de facto standard in face detection systems. It proceeds at 15 frames per second when the face detection algorithm was implemented using a conventional 700 MHz Intel Pentium III [1]. The goal of this project is to study Viola Jones algorithm in order to find and overcome Viola Jones weaknesses or limitations. This report documents experimental results done in MATLAB and discussions when various categories of input images are tested. It was realised that one of Viola Jones main limitation is its difficulty in detecting people of Negroid race, especially the darker the person, the more difficult he is to be detected. Through experiments with some image processing functions, it was found that the accuracy rate of the Negroid race increased by 147.9% when using Contrast-Limited Adaptive Histogram Equalization technique.