Streamlining GPU-Based Image Processing with OpenCL Code Generation

Abstract

In recent years, graphics processing units (GPUs) have become a game-changer in the image processing industry. An adaptable option for GPU-accelerated image processing is OpenCL, an open standard framework for heterogeneous parallel computing. We describe an image processing library code generator that, given a few directives included in a shader's source code, generates a wrapper code including all the OpenCL API calls necessary before executing the shader, simplifying the library's maintenance. For all of the evaluated operators, the proposed library outperforms the OpenCV, CImg, and ITK libraries.