PyGPU: A high-level language for high-speed image processing

Image processing is an area with many computationally demanding algorithms. When implementing an algorithm the programmer has to make the choice of either using a high-level language, thereby gaining rapid development at the expense of run-time performance. Or, using a lower-level language having higher runtime performance, but also a higher implementation cost. In this paper we present PyGPU, an embedded language that enables image processing algorithms to be written in the highlevel, object-oriented language Python. PyGPU functions are compiled to execute on the graphics processing unit (GPU) present on modern graphics cards, a streaming processor capable of speeds more than a magnitude higher than those of current generation CPUs. We demonstrate a number of common image processing algorithms, showing how these can be implemented succinctly and clearly using high-level abstractions, while at the same time achieving high performance.