A Fast and Accurate Faces Localization using Gradient Method

Georgy Kukharev, Pawel Masicz, Piotr Masicz
Technical University of Szczecin
Faculty of Computer Science and Information Technology
Zolnierska 49, 71-210 Szczecin
Poland

e-mail: pmasicz@wi.ps.pl

http://www.wi.ps.pl

Abstract

Detecting faces in images with complex background is a difficult task. To solve this problem we decided to use modified gradient method with oval object detection. This method is very fast and our improvements give better accuracy than it was in original gradient methods. Our system was tested using pictures delivered from digital cameras, camcorders and television. On those various sources differences in head size, lighting and background are considerable. During experiments we achieved very good results – 93-99% of correct face localization. We also implemented our method in C++ - it works well and fast.