Highly realistic video communication, which provide feel of existence of remote users and objects, is expected to be a next generation communication using broadband connection over the internet. To achieve such system, it is important to reproduce not only high quality image but also sense of directivity depending on user's position. However, viewpoint image synthesis is usually not easy for live action. In this paper, an approach for viewpoint image synthesis using multi-camera images and approximate depth information will be demonstrated. It can provide smooth motion parallax with high speed processing and high image quality.