48. A REAL-TIME FACE DETECTOR BASED ON SKIN COLOR AND HISTOGRAM OF GRADIENTS

Department: Computer Science & Engineering
Research Institute Affiliation: Graduate Program in Computational Science, Mathematics, and Engineering (CSME)
Faculty Advisor(s): Yoav Freund

Primary Student
Name: Sunsern Cheamanunkul
Email: scheaman@ucsd.edu
Phone: 858-336-3225
Grad Year: 2012

Abstract
Face detection has been around for many years. It is used in a wide range of applications such as video surveillance, or, recently, auto-focusing in digital cameras. The existing face detectors perform well in general. However, they fail to deliver the detection quality that is needed by the Automatic Cameraman system, which is installed on the 4th floor of the CSE building. The system requires a face detector to be able to run in real-time and to have the false positive rate as small as possible, while the false negative rate is not too high. In this work, we present a real-time face detector based on skin color and histogram of gradients that can fulfill the requirements of the Automatic Cameraman system. By using Integral Image technique, we are able to process color and gradient information and use them as detection features in real-time. For training, we use boosting on data collected by the Automatic Cameraman system. This allows us to train our face detector specifically for the system. Moreover, we utilize active learning to reduce the number of labeled examples needed in training. We can achieve reasonable performance using less than 500 labeled examples. Although the face detector is still in the training process, the latest results from our detector is already comparable to those from the OpenCV face detector.

« Back to Posters or Search Results