SegTC: Fast Texture Compression using Image Segmentation

Abstract

Fast, high quality texture compression is becoming increasingly important for interactive applications and mobile GPUs. Modern high-quality compression formats define sets of pre-existing block partitions that allow disjoint subsets of pixels to be compressed independently. Efficient encoding algorithms must choose the best partitioning that fits the data being compressed. In this paper, we describe a new method for selecting the best partition for a given block by segmenting the entire image into superpixels prior to compression. We use the segmentation boundaries to determine a partitioning for each block and then use this partitioning to select the closest matching predefined partitioning. Using our method for BPTC compression results in up to 6x speed-up over prior methods while maintaining comparable visual quality.