Image Edge Detection Using Canny and Sobel Techniques

Abstract

Finding edges is an important part of many image processing tasks, such as medical imaging, recognizing biometrics, and finding objects. But there isn't a single algorithm that works best in every case. We compared the Sobel and Canny operators on four standard grayscale benchmark images (Camera, Coins, Moon, Page) using five different metrics: runtime, edge pixel count, edge density, connected components, and average component area. Sobel was faster (8.10 ms vs. 13.28 ms) and made edges denser (8.93% vs. 6.59%). However, it also made almost twice as many connected components (469.0 vs. 217.5), which is a sign of fragmented edge maps because it is sensitive to noise. Canny always made edges that were thinner and clearer. This was especially true for the picture of the Moon (105 vs. 513 parts). Sobel is better for tasks that need speed, while Canny is better for tasks that need high accuracy, noise resistance, and edge continuity in all cases, one algorithm is not better than the other.