Research On The Method Of Image Edge Detection

Edge is the most basic feature of images, which includes the most part information of images. Edge detection is widely used in image analysis and processing such as feature escription, image segmentation, image enhancement and pattern recognitionetc,and has turned into a hot spot in research on image processing and analysis technology. So far, many algorithms have been presented in edge detection field. Despite all this, the edge detection of digital image has not been fully solved. The noise generated during the production and transformation of the images may result in false edge and undetected edge. Owing to the restriction of snapping environment and condition, there may be some irrelevant interferences. It is the major challenge in image processing to improve the accuracy and the signal-to-noise ratio of edge detection algorithm, thus making the algorithm an emphasis of professional study, also we have to work towards it.The main content of this dissertation is described as follows.1.Digital image processing and its applications are introduced, then, the background and the significance of the image edge detection technique are elaborated, next to this, some exist problems of the image edge detection are discussed.2.Some classical edge detection algorthms such as Sobel, Roberts,Deriche, Canny are discussed.Theory analysis and experiments are carried out to compare their advantages and disadvantages.3. Although there are many traditional methods for edge detection, most of them are sensitive to noise, especially some methods based on zero crossings, so that the image edge can't be detected clearly.This article presents a noisy image edge detection method, which is based on zero crossing. This method first smoothes the image and calculates gradient, then calculates the second derivative of the gradient image by the new recursive operator, and finds out the zero crossing points by rows and columns respectively, last merges all the zero crossing points and gets the image edge.The experiment results show that this method not only gives good edge for noisy image, but also spends less computation and time by recursive implementation of filtering operators.4.Considering the problem of undetected thin edge and delocalization in the edge detection technology, this article presents a edge detection method, which is based on the dual-threshold nonlinear derivative operator. Firstly, gray image is used to calculate the right and left derivative. Secondly, dual -threshold is used to injust the right and left derivative to remain the meaning edge information. At last, the two derivatives are merged to get the image gradient. t₁ can control the ability of reducing noise; t₂ can ensure detection of the one pixel width edge lines; and the nonlinear derivative scheme can solve the problem of delocalization. By comparing and analying it with the traditional discrete gradient operators in the experimental demonstrations, we find that this operator is not only with advantages of simplicity and higer accuracy of detection, but also get better edge image and higer SNR in the case of no any smootning.5.This article studies the application of the universal gravity in the field of edge detection and presents a gravitational edge detection method, which is based on the nonlinear filtering operator. Firstly, gray image is used to calculate the nonlinear gradient value of every pixel.Then construct a normalized function, whose independent is the nonlinear gradient value.Finally, the function value is used as the gray value of center pixel to do gravitational edge detection. By comparing and analying it with the traditional edge detection operators and gravitational approach of Sun et al in the experimental demonstrations, the proposed approach has better accuracy edge location and get better edge image for various noise images.