Abstract Recently a new type of coding method, called intelligent image coding, or more specifically model-based image coding, has attracted much attention as a basis for future visual communication services. The model-based image coding can be constructed by preparing a 3-D-shape model of an object on both the transmitting and the receiving sides, and by transmitting only information about movement and changes in the shape of this model. This paper discusses the model-based image coding in relation to moving facial images from two major objectives: extraction of motion information from an input image sequence and synthesis of an image sequence with a limited amount of motion information. First, the concept of intelligent image coding and the basic outline of model-based image coding are introduced, and then the procedures for processing the mouth and eyes are discussed. On the transmitting side, a feature-extracted image is obtained by applying a thresholding operation to an input image, and characteristic points which represent the basic shapes of the mouth and eyes are detected on it. On the receiving side, the shape of the 3-D model is modified according to the above results obtained on the transmitting side, and a realistic image is reproduced by assigning appropriate luminance and chrominance information to this model. A method of estimating the 3-D motion of the head is also briefly discussed. Experimental results demonstrate the feasibility of a new type of coding system based on the model-based image coding described in this paper.