The Encoding Algorithm And GPU-Based Optimization In Image Compression Of Digital Film

JPEG2000 combines excellent compression performance and code stream with good scalability. Digital Cinema Initiatives (DCI) has recommended it as the encoding algorithm to be used for digital distribution of motion pictures. However, due to its complexity and time-consuming of algorithm, it hasn’t been widely used in digital film system of high real-time requirement. In order to make a better encoding technology development of digital film, our work focus on the optimization of encoding algorithm to improve the encoding efficiency.At present, the development speed and computational capabilities of Graphics Processing Unit (GPU) have exceeded more than that of CPU. Especially, Compute Unified Device Architecture (CUDA) proposed by NVIDIA in 2007 makes it easier and more effective to process massive data in parallel by using multiprocessors. Our topic has selected CUDA as the hardware platform because of its computing power and the algorithm characteristics of image encoding.The main work in our research is the optimization of four parts of the whole encoding algorithm which take up about 69.9 percents of the total encoding time. Firstly, we make an optimization of preprocessing of image and quantization by the proper allocation of resources and tasks. Secondly, by the operation of transposition we efficiently realize the GPU-based Discrete Wavelet Transform. Thirdly, we have redesigned and achieved a new parallel Bit-Plane Coding algorithm including pass prediction, significance states update and three pass coding which is suitable for CUDA.Our experimental results demonstrate that it is more than 40 times speedup in Discrete Wavelet Transform and about 4 times speedup in Bit-plane Coding compared with the original implementation on the CPU. And also the Preprocessing of image and Quantization has got respectively 33.4 and 20 times performance improvements.In addition, according to CUDA and Message Passing Interface (MPI) hybrid paradigm, we have designed and realized the calculation system effective for the image sequences encoding of digital film by using multiprocessors including CPUs and GPUs.