PARALLELIZATION OF EDGE DETECTION ALGORITHMS BASED ON GPU

Abstract

Fhe biggest challenge in the field of image processing is that intensive computation power is required to achieve high accuracy and real-time performance. Real-time image processing of video frames is difficult to attain even with the most powerful modem CPU. High-resolution video capture devices and increased requirements for accuracy make it even harder to realize real-time performance. One way to achieve this is to use GPU. Recently, GPU has evolved into an extremely powerful computation resource. Thus, they can exploit the single instruction multiple data (SIMD) architecture and be effectively parallelized on GPU. Also CUDA (Compute Unified Device Architecture), which is a novel technology of general-purpose computing on the GPU, makes users develop general GPU (Graphics Processing Unit) programs easily. In this dissertation, different edge detection techniques are implemented on CUDA and their results are compared.