基于异构平台的自适应图像去马赛克的OpenCL加速

摘要

基于异构计算概念，使用 GPU 和 OpenCL 加速了一个高复杂度的自适应图像去马赛克算法，并在 AMD Bald Eagle 和 FirePro W8100组成的异构计算平台上完成了功能和性能测试。实验结果表明，该异构平台能取得良好的图像重建效果， W8100处理图像的速率超过了100 f/s ，每帧图像有1920×1080个像素，证明异构计算平台及 OpenCL可满足医疗、网络监控等应用领域对高帧率、高清图像影像的需求。%This paper describes a GPU-accelerated implementation of adaptive image de-mosaicking ( a common digital image pro-cessing application ) to demonstrate massive computation power of the many-core architecture . Acceleration strategies are briefly de-scribed to properly map the real-world applications to the GPU for high performance computing . Experimental results show that the developed OpenCL implementation can leverage the AMD FirePro W8100 , which achieves a high throughput rate up to 100 frames/s , each having 1920-by-1080 pixels with RGB values . This research work integrates this GPU processor and the AMD Bald Eagle ( an embedded APU processor ) to form a heterogeneous computing platform , and uses this acceleration example as a case study to demonstrate that heterogeneous system architecture is well-suited for high-throughput , high-definition applications in many fields , such as security and medical imaging .