Study On Sub-pixel Image Edge Detection Method

The purpose of the image as carrier is to obtain useful information from the image information, and then make a right deal to detected object. In this process, accurately extracting image edges is very important, because the image edges contain a lot of useful information for the target object and important relevant parameters. Therefore, edge detection technology play a very important role in the computer and machine vision detection system, and has an important impact to subsequent image processing, for example, object registration, object detection and recognition, image segmentation, region contour extraction, object measurement. However, with the development of science and technology, pixel-level edge detection accuracy cannot meet the needs of the actual production and application, therefore, the sub-pixel edge-based detection technology has become a practical significance research. This thesis studies mainly sub-pixel-based edge detection methods, the main work is as follows:(1) In this paper, the pixel-level based edge detection methods and the sub-pixel edge detection methods are respectively briefly introduced.(2) Based on edge step model, traditional pixel edge detection operator-Sobel operator is used for edge detection, and edge parameters (position, direction) are calculated using the two edge templates of pseudo-Zernike moments in the position of the detected edge pixel level. Taking into account the actual edge model, we take advantage of the edge error compensation model to modify edge parameters calculated by step edge model; thereby accuracy of detected edge parameters is increased.(3) Different from current popular continuous sub-pixel edge detection models, we use discrete data to establish the equation edge in order to detect sub-pixel edge, the specific method is:first edges are detected by using traditional pixel-level edge operators such as Sobel, Canny, etc., and then detected edge position as the center, set up a 3 × or n × 3 area, six equations are established by using one-dimensional gray moment in this area, in order to obtain four sub-pixel edge parameters (position, direction, the target gray value, the background gray value), and then the sub-pixel edges are detected.