3-D motion estimation for model-based image coding
T. Fukuhara, A. Umahashi, Tsuyoshi MURAKAMI · 1992
Motion estimation is one of the most important techniques for model-based image coding. Accurate and robust estimation of motion is required in lots of application fields, such as detection, recognition, tracking of moving objects, and face-to-face visual communication. These communication systems require very accurate motion estimation of moving objects, especially the human head. In the paper, the authors present a new motion estimation method the human head using a 3-D shape model and 3-layer neural network for model-based image coding. After the existing methods of 2-D motion estimation are reviewed, the authors propose a method of 3-D motion estimation using a neural network and 3-D shape model. The neural network consists of three layers. Input layer represents 2-D motion vectors of feature points, while output layer represents 3-D motion parameters.