Abstract
In the thesis, a new feature set which is composed with Gabor feature and Haar-like feature named hybrid feature set is proposed. The goal of this thesis is to create an automatic face detection system which is robust to pose and head motion. Our face detection system consists of two modules. The first module searches the potential face regions by using skin color detection and segmentation procedures. The second module selects the features of the scanned image. This system can be used in different sizes, varying poses, different expressions, and defocus problems. From the experimental results, we find that we can have a better system performance comparing with the classifier using only single type weak classifier. Our face detection system can detect the faces with rotation angles from -90? to 90? with an average correction rate about 93%-95%. Our system can also be applied for facial expression recognition with high correction rate. We also do the real-time test of the system by using AMD 3000+ CPU and the image size is 320*240 pixels. It requires 300~350 ms to detect a face and 50~60 ms for tracking. Therefore, our system is more robust than other proposed face detection system and can be widely used.