CUDA-based acceleration of post deblocking filter

In this paper, an accelerated post deblocking filter algorithm based on CUDA (Computer unifiled device architecture) technology is proposed. The algorithm is used to remove block artifacts and improves the visual quality. It is computationally intensive and usually requires high speed processors to run in real time. A commonly efficient way can be obtained using the DSP (Digital Signal Processing) or other hardware resources. Here, we put forward a software solution method. Using NVIDIA´s CUDA technology, we save more hardware resources and make up the low programmability of hardware. CUDA technology assists GPU (Graphics Processing Unit) to work for CPU for large computation. In the experiment, a frame was picked up from JM (Joint Model) decoder stream and filtered by the improved algorithm. Without changing subjective quality, the result shows that the processing time saved 5.69 times than CPU.