A Research About Video And Image Processing Based On Heterogeneous Computing Technology

In recent years, wide-ranging developments in cloud computing technologies accelerate the appearance of new services on the cloud computing platforms. As a new type of cloud services, cloud desktop and cloud game require computing platforms use complicated algorithms to process a mammoth of video and images in real-time. However, such requirement is impossible to be fulfilled on the current cloud computing systems because the existing cloud systems generally rely heavily on the x86 CPU cores to complete the given computation workloads. These systems are not efficient in measures of performance and energy consumption to support the new cloud services. Accordingly, these new services have not entered into the market yet. In that case, it is of great significance and value in real-world applications to investigate a cost-effective solution to address the issues related to the data-intensive and computation-intensive cloud services.In this dissertation, heterogeneous computing is proposed to improve computational performance because its architecture consists of different types of general-purpose cores(e.g. x86 CPU, GPU, FPGA) as well as specific-purpose ASICs(e.g. VCE and UVD in the AMD APU processors). These new micro-processors provide suitable supports for different work-loads for the sake of high efficiency in performance and energy-consumption.Image capturing, image processing and video decoding are studied in this dissertation. It is proposed to use computing resources in heterogeneous computing environment, such as GPU and UVD, for accelerating video and image processing. This acceleration can achieve real-time processing for video and image with high quality. Three functions are designed and implemented in this dissertation: 1) Video source image capturing accelerated by GPU parallel computing in image transformation; 2) Image processing accelerated by GPU parallel computing with OpenCL; and 3) HEVC decoding with UVD of new generation. A contrast experiment between the proposed method and the conventional solution is conducted in detail to present the distinction performance. Experimental results demonstrate that video and image processing accelerated by heterogeneous computing technology in this dissertation performs great on AMD heterogeneous computing platform. In contract with serial implementation based on x86 CPU, its computational acceleration is remarkable. It is sufficient for the requirement of real-time video and image processing with high quality.