A new intra-residual compression method for BC7/ASTC texture formats is described, which uses neighboring already-coded blocks to predict the current block's weight grid, then applies a 2D DCT/DST transform to the residual, quantizes the coefficients, and codes them. This extends the earlier XUASTC LDR absolute weight grid DCT approach (JPEG-like) to a prediction-based scheme (WebP-like). For BC7, the predictor is a dequantized 4x4 [0,64] weight grid subtracted from the current block's weight grid before DCT coding; ASTC requires resampling the predictor to match the current block's weight grid resolution.
35 Impressions