Image-processing methods on general-purpose graphics processors with parallel architecture

This paper discusses the principles of massively parallel computations for digital signal processing by means of NVIDIA CUDA technology, using as examples such operations as the inversion of image brightnesses, gamma correction, and the Sobel operator. The main methods of digital image processing using massive parallelism of the computations on a graphics processing unit are evaluated. These methods are implemented on the central processing unit and the graphics processing unit and are compared in terms of such parameters as the time to carry out the processing, the size of the images that are used, and the size of the memory blocks used by the CUDA architecture. (c) 2012 Optical Society of America.