S3Dc: A 3Dc-based Volume Compression Algorithm

Abstract

Volumes acquired for medical purposes are continuously increasing in size, faster than graphic cards memory capacity. Large volumetric datasets do not fit into GPU memory and therefore direct rendering is not possible. Even large volumes that still fit into GPU memory make frame rates decay. In order to reduce the size of large volumetric models, we present a new compression scheme. In this paper we present S3Dc, a lossy volume compression algorithm suitable for scalar values. It is inspired in hardware-accelerated 3Dc normal compression technique. S3Dc allows us to compress the volume in CPU up to a 4:1 or 8:1 ratio, while still yielding good quality results. We provide details on the compression scheme and show how to render directly from a S3Dc compressed texture. Furthermore, we analyze the image quality theoretical error and the average error with several images in order to assess the results.