Multi-view Face Detection Using Six Segmented Rectangular Features

This paper presents a multi-view face detection system which combines skin color detection and adaptive boosting (Adaboost) algorithm. The aim of this combination is to satisfy the accuracy and speed, the two important characteristics of real time face detection. The second contribution of this paper is a new type of rectangular features for face detection, represented in a 2 X 3 matrix form. With these new features the training time becomes significantly very short: five times faster than using the traditional feature set. The experimental results demonstrate the effectiveness of our method in detecting profile and rotated faces over a wide range of variations in color. The method detects 97.5% of positive faces while 5% is declared as false positive. The system also detects the occluded faces as well as lean faces and rotated faces.