Title :
Fast 2D and 3D image processing with OpenCL
Author :
Daniel Oliveira Dantas;Helton Danilo Passos Leal;Davy Oliveira Barros Sousa
Author_Institution :
Universidade Federal de Sergipe, Departamento de Computaç
Abstract :
Image processing with GPUs requires the use of an API like OpenCL or CUDA. A higher level library that hides these APIs is a better option if the programmer does not need to fine tune or implement his own image processing operations. Among recent libraries that support OpenCL, there are Open-CLIPP and OpenCV, although none of them supports 3D images. Using OpenCL, however, is not as simple as programming CPUs, and many API calls are needed to prepare the environment before calling the image processing function (shader). In this article we describe a library with a code generator that, from a few directives merged in a shader source code, generates a wrapper code with all the OpenCL API calls needed before calling the shader, simplifying, thus, the maintenance of an image processing library. The proposed library performance is better than OpenCV, CImg, ITK libraries for all the tested operators.
Keywords :
"Graphics processing units","Libraries","Context","Image processing","Three-dimensional displays","Random access memory","DICOM"
Conference_Titel :
Image Processing (ICIP), 2015 IEEE International Conference on
DOI :
10.1109/ICIP.2015.7351730