Face Localization and Versatile Tracking in Wavelet Domain

Automatic initialization and tracking of human faces is one of the first steps in designing interactive multimedia web applications and security systems. In this paper, we present an effective method for face tracking in color image sequences in wavelet domain. We have used the chrominance component of nonlinear YCbCr color model to segment face-candidate regions in the first frame of the image sequences. In our proposed method, after the face detection stage, the face tracking is carried out by an efficient block-based matching algorithm in wavelet packets domain. A block transform coding technique is employed to reduce the spatial redundancy. By using the amplitude of coefficients in the best basis tree expansion of the wavelet packet transform, the feature vectors of the face reference block are extracted. The best matched block has determined the new position of the face. Robustness against face deformations, lightening changes, occlusion, and presence of noise, as well as the performance of its subsequent stages are demonstrated in various experiments