An Edge Detection System for Polluted Images by Gaussian, Salt and pepper, Poisson and Speckle Noises

Due to the problems which noises may cause in receiving, transmission and processing images, and the problems that traditional edge recognition methods such as: Canny, Zerocreoss, Log, Roberts, Prewitt and Sobel have in relation to these noises, especially Gaussian, Salt and Pepper, Poisson and Speckle, we decided to design a system which is not only an edge detection method, but in the case, facing different noises, is able to indicate least of sensitivity. This system comprises three section, which initially divide the RGB image to its constituent colours, and then applies a filter to them in order to make it smooth (without changing the real position of the edges). In the next section, a new filter is employed to display the edges, and in the last section, a post-process will takes place on the output binary image for eliminating superfluous spots. The acquired results of comparison of this system with traditional edge detection operators indicates the high efficiency and robustness of it, against different noises. Also in order to statistically validation, we benefit from PSNR, MSE and SSIM factors. Even statistical results indicate the high similarity of noisy edge detected images and non-noisy edge detected images in the proposed system.

کلیدواژه ها:

Image Processing ، Edge Detection ، Noise ، Binary Image ، Filter