3D Wavelet Lifting Video Encoder And Optimized With GPU

In JPEG2000, the compression method which is based on wavelet transform is a branch of field of video encoding. It is suitable for the movies on high-definition. But without using the interframe prediction completely, it is not efficient when it compressed digital cinemas only use intra-coded. In order to solve this problem, we propose a new video compression technology. The core of this technology is three-dimensional wavelet transform. Then we parallel optimize it.In recent years, the development of GPU has is faster than CPU. The gap of performance between CPU and GPU is growing quickly. When NVIDIA released CUDA in2007, we can develop parallel programs without using any graphics API. This would resolve the problem that GPU programming is too hard. Our research uses CUDA as the hardware platform. We can use its outstanding parallel ability to optimize3D wavelet algorithm.Our research’main work is as follows. We have added decomposition of time domain into the traditional space wavelet decomposition, with which a3D wavelet encoding method comes out. We also add motion compensation into this method, so the temporal redundancy can be removed and the image quality can be protected. Then, we optimized the2D wavelet lifting algorithm,3D wavelet lifting algorithm and motion estimation algorithm with GPU. All of them are used in our encoding system. In addition, in order allow the results to get a full comparison, our research has used OpenCL (Open Computing Language) to optimize the serial program on CPU. Our experimental shows that, compared to the serial program, we have got94and8times speed up in2D wavelet lifting algorithm, got56and11times speed up in3D wavelet lifting without motion compensation, got11and9times speed up on the3D wavelet lifting with motion compensation.In addition, we have presented a distributed system with GPU and CPU. This system can make good use of the computing ability of CPU and GPU, and realizes the computing platform which using multiprocessors including CPUs and GPUs. With it the entire algorithms can be more efficient.