The Research On Methods Of Edge Detection

Edge is one of the basic characters of an image, which offers people important parameters to describe and recognize objects in image processing. Edge detection is widely used in the recognition, segmentation, intensification and compression of the image. In addition, it is often applied to the high-level domain such as computer vision and pattern recognition.The research on edge detection has a long history, and there have been lots of algorithms proposed for it. But all of the algorithms for edge detection still have some drawbacks and can not detect the optimal edges in some cases. It is hard to propose a general method of edge detection applied to all cases. So the main research orientation of edge detection is to make improvements to existing methods or to find new methods for edge detection with specific application requirements.Images obtained from real-world scenes are generally buried in noise. How to detect edges reliably and accurately in the presence of noise has remained an important issue in the field of edge detection. The classic edge detection methods have some drawbacks in image denoising. The main idea of resolving this problem is to set a threshold, then compare the threshold with the high-frequency components of an image to remove the noise. So the choice of the threshold is the key of image denoising. The threshold of traditional methods is obtained by experiments, and there is no general method to determine it. A new method of edge estimation by the Maximum a Posteriori (MAP) is presented in this paper. It proves how to choose the optimal threshold in theory.Laplacian operator is a second derivative operator, which can detect the edge by the zero-crossings in the second derivative. The Laplacian operator can detect the intensity abrupt change and get better result in edge localization, but it is sensitive to noise. In this paper, based on the Laplacian operator, a model is introduced for making three detectors. Then, by the optimal edge-matching filter, the multistage median filter, the max/min median filter, three detectors are obtained. Experimental results illustrate that the method using max/min filter has a good noise removing performance and can provide a satisfactory result in detecting edges. The paper is organized as following five parts. Firstly, Section 1 gives an introduction of the background, significance and development of the field of edge detection. And Section 2 introduces several classic edge detectors and gives edge detection results of these methods and offers some relative analysis and comparison. Then Section 3 offers the Sobel operator and the improvement to it, and introduces a new method for thresholding to removing the noise. In Section 4, a new edge detection model, the Laplacian operator-based edge detection model is proposed. Three new edge detectors based this model is introduced: the optimal edge-matching filter-based edge detector (OED), the multistage median filter-based edge detector (MED) and the max/min median filter-based edge detector (MMED). In addition, experimental results and analysis of the performance of three new edge detectors are presented. At last, the paper ends with a conclusion and some further research plans mentioned in Section 5.