Parallel BTF Compression with Multi-Level Vector Quantization in OpenCL

Abstract

Bidirectional Texture Function (BTF) as an effective visual fidelity representation of surface appearance is becoming more and more widely used. In this paper we report on contributions to BTF data compression for multi-level vector quantization. We describe novel decompositions that improve the compression ratio by 15% in comparison with the original method, without loss of visual quality. Further, we show how for offline storage the compression ratio can be increased by 33% in total by Huffman coding. We also show that efficient parallelization of the vector quantization algorithm in OpenCL can reduce the compression time by factor of 9 on a GPU. The results for the new compression algorithm are shown on six low dynamic range BTFs and four high dynamic range publicly available BTF samples. Our method allows for real time synthesis on a GPU.