Implementation of image processing algorithms based on GPU

摘要

The most perspective direction in the image processing became the use of graphics processing units (GPU). The main advantage of it is the exceptional dedication to work with 2d and 3d graphics, respectively processing large volumes of similar data. The consequence of this was the emergence of a single graphics card on the board of several stream processors responsible for performing the same type of command. However, a major breakthrough in the use of the GPU is the emergence of the possibility of programming the stream processors (vertex and pixel) image processing. The special feature of the group of processors are parallel processing, i.e., for a pixel processor unit is one piece or texel. For example, modern GPU carry on board more than 1,000 stream processors, each type of which can significantly improve performance by parallel use of all processors. Work related to the parallelization takes on the GPU, the engineer must identify only the basic algorithm for a single processor.