3D Mosaic from Images

Abstract

This paper presents a tool for 3D object mosaic. Given a set of background removed input images, we first compute a 3D reconstructed volumetric model body using shape from silhouette. The granularity of a volumetric body is the user input. Voxel center coordinates, voxel color, and surface normal of the voxel are computed for 3D mosaic. The voxels of reconstructed volumetric body are replaced by primitive shapes such as sphere, cylinder, cone, etc. We call this process as a 3D mosaic. The background-eliminated input images may contain information on body parts supplied by a user. Using information on body parts, only a part of 3D reconstructed volumetric body is replaced by a new shape while the rest of body retains voxel information. The surface normal values are used for primitive shapes with direction such as a cone. 3D mosaic can be used for emphasizing or deemphasizing a part of 3D reconstructed model body, similar to the function of a 2D image mosaic. Emphasizing and deemphasizing is done by resolution, surface normal, size of body parts, color and/or shape of the 3D primitive object.