GPU-driven recombination and transformation of YCoCg-R video samples

Common programmable Graphics Processing Units (GPU) are capable of more than just rendering real-time effects for games. They can also be used for image processing and the acceleration of video decoding. This paper describes an extended implementation of the H.264/AVC YCoCg-R to RGB color space transformation on the GPU. Both the color space transformation and recombination of the color samples from a nontrivial data layout are performed by the GPU. Using mid- to high-range GPUs, this extended implementation offers a significant gain in processing speed compared to an existing basic GPU version and an optimized CPU implementation. An ATI X1900 GPU was capable of processing more than 73 high-resolution 1080p YCoCg-R frames per second, which is over twice the speed of the CPU-only transformation using a Pentium D 820