Abstract: Edge detection is a highly parallel algorithm with great computation. It is difficult to increase the speed of the algorithm by CPU to satisfy the real time application. Aiming at the computcintensive character of image edge deteclion, this paper analyzed some methods of edge detection based on CPU, using the programmer friendly CUDA framework, and proposed a method based on GPU, to realize the image edge detection. The efficient architecture of GPU was introduced firstly. Then, two representative image edge detection algorithms, Roberts and Sobel, were implemented on GPU. At last, using the same market price level CPU and GPU as hardware platform, and using various resolution images as test data, compared the computational efficiency of GPU and CPU. Numerical experiments show that the speed of the algorithm can be improved by up to more than 17 times compared with CPU-based implementations,with the same processing results. It proves that the GPU is practical for some applications of image processing.

Key words: Image processing, Edge detection, GPU, CUDA