Face Representation and Tracking Using Gabor Wavelets Network

Many approaches to human gesture recognition exist. This work presents one of the approaches. Especially the problems of human face representation and tracking in video sequences are discussed. A method, where a discrete face template is represented by linear combination of continuous 2D odd-Gabor wavelet functions (Gabor Wavelet Network-GWN) is proposed. The weights and 2D parameters (position, scale and orientation) of each wavelet are determined optimally. Using this representation, an effective face tracking method is achieved that is robust to illumination changes and deformations of the face image such as smile. Possibilities of cooperation between a GWN and others methods for robust facial expressions recognition are also discussed, for example: PCA (Principal Component Analysis) and parameterized models of optical flow.