A multi-agent based pose and scale invariant human face recognition system called iJADE Face Recognizer is presented. We focus on how neural networks are applied on face detection under cluttered scenes, and committee network handles detection of multi-pose faces. We also investigate how Gaussian mixture model of skin tone can narrow down the region of interest in complex image on detection. In feature extraction on faces, Gabor feature vector derived from Gabor wavelet representation of faces is adopted, which is robust to changes in illumination and facial expression, and we utilize template and feature-based face recognition methods in order to improve the recognition rate. In face identification process, we make use of agent technology to increase the scalability and efficient of the system. By using these techniques, we develop an accurate and efficient recognition system with invariant to different conditions on human faces under uncontrolled environment.
展开▼
