A compact representation for scanned 3D objects

We consider the representation of densely sampled scanned 3D objects. Scanning physical objects can be an efficient way of obtaining natural looking input for computer graphics image generation. Scanned meshes however require large amounts of storage. For applications such as computer graphics, preserving the look of objects rather than precise measurements is critical. We seek a representation that preserves characteristic features at both low and high spatial frequencies. Our proposed representation consists of a simplified base mesh and characteristic detailed patches. The detailed patches contain a spatially dense sampling of mean curvature on a small region on the object surface. A large initial set of patches covering the object is segmented into sets of patches using k-means clustering. Each cluster of patches is then compactly represented with Gaussian distributions. We present comparisons of a series of original scanned objects and the objects reconstructed from the compact representation.