Parallel Implementation of Souvola's Binarization Approach on GPU
Binarization is widely used technique in many of the image processing applications. Fast algorithms are needed for fast and efficient image processing systems. Many algorithms of image processing and pattern recognition have recently been implemented on Graphic Processing Unit (GPU) for faster computational times. GPUs are most prominent hardware in utilizing parallelism and pipelining than general purpose CPUs. Moreover, Speed, programmability, and price become it more productive. In this paper, the authors proposed a parallel implementation of well known Sauvola's local binarization algorithm for Optical Character Recognition systems. In this experiment, they achieved a computational speedup of parallel implementation on GPU 20.8x times faster than implementation on CPU. The speedup results of GPU are promising.