Parallel Implementation of Otsu’s Binarization Approach on GPU

Fast algorithms are important for efficient image processing systems for handling large set of calculations. To speedup the processing, parallel implementation of an algorithm can be done using Graphics Processing Unit (GPU). GPU is general purpose computation hardware; programmability and low cost make it productive. Binarization is widely used technique in the image analysis and recognition applications. In this paper, we investigate the accuracy and performance characteristics of GPUs on well known global binarization Otsu’s approach for Optical Character Recognition systems. The main goal of this research work is to make binarization faster for recognition of a large number of document images on GPU. The algorithm is implemented using Compute Unified Device Architecture (CUDA). Experimental results show that parallel implementation achieved an average speedup of 1.6x over the serial implementation when running on a GPU named GeForce 9500 GT having 32 cores. Otsu’s method is also evaluated using PSNR, F-measure, NRM, and IND evaluation measures. General Terms Document Analysis and Recognition, Image Processing, Pattern Recognition.