A new model based compression scheme of face images is proposed as a combination of context-based and classic compression schemes. Deformable templates are employed to locate salient features of face images whose residuals are refined in a further residual vector quantization. Normalized face images are decomposed into subbands by using bi-orthogonal spline filters. Subbands and their first degree residuals are vector quantized sequentially in a pyramidal fashion. Wavelet coefficients are zonal truncated.
展开▼
