Skip to Main Content
Many of the basic image processing tasks suffer from processing overhead to operate over the whole image. In real time applications the processing time is considered as a big obstacle for its implementations. A High Performance Computing (HPC) platform is necessary in order to solve this problem. The usage of hardware accelerator make the processing time low. In recent developments, the Graphics Processing Unit (GPU) is being used in many applications. Along with the hardware accelerator a proper choice of the computing algorithm makes it an added advantage for fast processing of images. The Cellular Neural Network (CNN) is a large-scale nonlinear analog circuit able to process signals in real time . In this paper, we develop a new design in evaluation of image processing algorithms on the massively parallel GPUs with CNN implementation using Open Computing Language (OpenCL) programming model. This implementation uses the Discrete Time CNN (DT-CNN) model which is derived from originally proposed CNN model. The inherent massive parallelism of CNN along with GPUs makes it an advantage for high performance computing platform . The advantage of OpenCL makes the design to be portable on all the available graphics processing devices and multi core processors. Performance evaluation is done in terms of execution time with both device (i.e. GPU) and host (i.e. CPU).
Date of Conference: 25-27 July 2011