GPU-accelerated computation for texture features using OpenCL framework

Texture Features introduced by Haralick in 1973 which rely on computing the so-called Gray Level Co-occurrence Matrix (GLCM), are being used extensively by many applications to understand and enhance images acquired from various scientific contexts. The main limitations of these features are their high computational costs pertaining to memory usage and processing time. In this paper a Graphics Processing Unit (GPU)-based parallel implementation for computing the GLCM along with the features is introduced using the Open Computing Language (OpenCL) framework to be exploited in Segmenting Synthetic Aperture Radar (SAR) images. An optimized, sequential C-language version of the software computing GLCM and features is presented then our parallel implementation is explained pointing out the speed up achieved, which hits a factor of 340x compared to the sequential implementation.