Abstract
A model‐based image coding system basically consists of an “analysis” process at the transmitting side and a “synthesis” process at the receiving side. Images are coded by transmitting only information about changes in shape of a three‐dimensional (3‐D) model of an object prepared at both the transmitting and receiving sides. The essential elements in the “analysis” process are recognition of the object from moving images, estimation of the object motion, and detection of changes in its shape.
This paper proposes a method of estimating the motion in 3‐D space of the head in model‐based image coding of moving facial images and confirms the validity of the method by computer simulations.
First, the basic concept of model‐based coding is described. A discussion about the treatment of motion information in moving images is performed from the viewpoints of 3‐D motion of the object and shape changes of object components, and the outline of the basic idea to detect the motion and shape changes is presented. Then the 3‐D motion estimation method using the 3‐D shape model of the object and motion vectors in 2‐D space is described in detail.
Computer simulations are performed on moving facial images, and the results show that the motion of the head in the input moving images is well reproduced as the motion of a 3‐D shape model.
Finally, the relation between precision of depth coordinates and estimation results is considered, and it is demonstrated that the proposed method is effective for estimating 3‐D motion parameters in model‐based coding of moving facial images.