Compression of Sequential Voxel Data for Sequence-based Animation

Abstract: This thesis presents a method for lossy compression of voxel data in an animation sequence. Uncompressed voxel animation can become very memory intensive and therefore need a more efficient method to store and render such data. To achieve this, a combination of methods were implemented in order to create a compact data structure, given the name Hyper Octree. This method first involves the use of a modified sparse voxel octree that is able to exploit spatiotemporal coherency between each consecutive frame in a sequence. The second method is a compression of colours using the DXT1 format in order to further compress the data. This format causes a loss in quality, so a novel approach of using HSV values to sort or reorder the colours in order to be in close proximity to each other in memory before compression. This format also causes redundant data to be generated, so the data is reduced as well to further compress the data. The result shows that the Hyper Octree was able to greatly reduce the memory consumption and the HSV sorting was able to preserve the quality of the colours moderately. Sorting colours with HSV appeared to be most effective depending on which component in the HSV colourspace was the most frequent in the voxel data’s set of colours.