Computer-generated holograms (CGHs), which are generated by simulating the recording process of a hologram
in a computer, are noted as an ideal three-dimensional (3D) display technology. However, with CGHs it
is necessary to create precise 3D model data based on objects that already exist, and it is difficult to do this.
To solve this problem, there has been much research on generating CGHs using multi-view images (MVIs).
MVIs make it possible to generate CGHs from real-existing objects in natural light. A method using ordinary
digital cameras resulted in high-resolution reconstructed images without the need for any special devices, but
with this method it is necessary to capture a huge number of images or to use a huge number of cameras to
ensure a sufficient continuous motion parallax. This is simply not realistic for the construction of 3D display
applications. In this paper, we describe a method of generating voxel models from captured images and then
using the MVIs obtained by the models to generate CGHs. We generate voxel models by SFS, determine
voxel value using the captured images, and render voxel models into MVIs. Using this method enables us to
arrange holograms at arbitrary positions in the range in which MVIs are generated correctly. We can also
obtain a sufficient continuous motion parallax by generating MVIs obtained from voxel models in spite of
capturing only a small number of images. Results of optical experiments demonstrated the effectiveness of
the proposed method.