基于异构平台的自适应图像去马赛克的OpenCL加速

An OpenCL implementation of an image de-mosaicking algorithm on a hybrid platform of heterogeneous system architecture

基于异构计算概念,使用GPU和Open CL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Fire Pro 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1 920×1 080个像素,证明异构计算平台及Open CL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 Open CL implementation can leverage the AMD Fire Pro 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.