Gaussian Filtering Implementation and Performance Analysis on GPU

Abstract

In image processing field, filtering process places vital role in order to extract high quality picture. The comparison demanding of the CPU operation is more and hence the performance of convolution in image processing takes more time. When compared to CPU the GPU may be a better way in accelerating image filtering. NVIDIA has developed a parallel computing platform known as CUDA (Compute Unified Device Architecture). CUDA is a programing interface which uses the parallel architecture for general purpose computing that suits for highly general purpose programing on GPU. This interface is known to be a set of library functions which could be coded as an extension of C language. In this paper the filtering is implemented with the help of two code languages the formal one is OPENCV and the other is the CUDA implementation. GPU modules are being included in the OPENCV library which contains all the GPU accelerated stuffs. NVIDIA supports the work on the module. The OPENCV GPU programing is written in CUDA and it benefits from CUDA ecosystem. The inputs are drawn with the CPU function and the filtering operation is done as the GPU function, again the results are copied into CPU and displayed. We observed that the GPU has the high speed up when compared with the CPU in image filtering operation.