Image filtering is one of the important preprocessing steps in image processing. Advancement in the technology has brought development in both spatial and frequency domain. In Spatial filtering, the filtering operations are performed directly on the pixels of an image. Spatial filtering includes various techniques like gaussian blurring, edge detection, denoising, etc. This paper concentrates on accelerating image filtering with Graphical Processing Unit (GPU), such that the speedup and performance of the filtering process can be enhanced. GPU is an efficient way to accelerate image filtering and uses NVidia’s Compute Unified Device Architecture (CUDA) technology for parallel computing. Traditional CPU can run only a few complex threads concurrently. GPU allows a concurrent execution of hundreds or thousands of simpler threads in parallel. This paper includes two filtering techniques, namely, gaussian blur filter and sobel edge detection filter for grayscale images on GPU using CUDA programming language.

Fundamental steps in Digital Image Processing are image acquisition, image enhancement, image analysis, image reconstruction, image restoration, image compression, image segmentation, image recognition, and visualization of image (Rafael et al., 2005). Image enhancement improves the image quality by applying various image filtering techniques in spatial domain and frequency domain. Spatial domain methods directly modify the image pixels to achieve desired enhancement in spatial domain. Frequency domain methods perform the enhancement operations to Fourier Transform (FT) of an image in frequency domain.

Image filtering techniques are classified into two domains: spatial domain and frequency domain (see Figure 1). Spatial domain includes two filters, namely, smoothing and sharpening filters.

Computer Sciences Journal, Image filtering, Parallel computing, GPU, CUDA, Performance Enhancement, Image Filtering, GPU, Graphical Processing Unit (GPU), Compute Unified Device Architecture (CUDA).