Research Of Ellipse Detection Algorithm Based On Arc Group

Ellipse detection is a basic task of image matching and recognition.image understanding and analysis.computer vision and othcr fields.Many researchers have studied a lot on ellipse detection,and have proposed a variety of ellipse detection algorithms.which can be divided into three categories:voting/clustering. optimization,method based on arc.The first two algoorithms directly detect ellipse based on the large number of discrete points in the image. With the increasing complexity of the image.the number of edge points increase,the number of noise, isolated points and independent paints also increase.This will affect the efficiency and accuracy of ellipse detection.Therefore,researchers have proposed some methods using are instead of discrete points as ellipse detection data.Since arc contains relative more accurate and complete geometry information than discrete points,it can avoid a large number of other unrelated edge points in elljpse detection.This kind of method has certain advantages on the efficiency and accuracy of ellipse detection.Based on the reseach and analysis of common ellipse detection algorothms,form boundary tracking,are extraction,are grouping,ellipse detection and other aspects this paper has discussed key issues relaied to ellipse detection algorithm based on are group and proposed solution(1)In order to eliminate image noise and isolated points,correctly connect adjacent edge points,obtain continuous,numerical smoothing boundary segment in local neighborhood,and as much as possible avoid wrong branch of boundary segment,this paper has designed a smooth boundary segment extracting method guided by the gradient direction of points.This method uses the direction infomation of points contained in recently tracked chain code to determine the search direction. and uses the gradient directiong of points to guide the track path of branch on the boundary,so it can achieve 1 to 2 breakpoints on the boundary across,and effectively reduec wrong choose on the branch of intersect boundaries.The effect is better than 8-connected boundary tracking algorithm.(2)In order to remove other adge points unrelated to ellipse detection in image. correctly extract arc needed, according to the geometric properties of ideal elliptic boundary without corner and inflection point, this paper has designed a segmentation-point detection method. This method uses the discrete curvature of edge points to determine corner points, and uses the convex and concave features of 4 adjacent points to locate inflection points, so it correctly realize segmentation of boundary. Then, with the features of elliptic arc. it can remove segments which do not constitute elliptical arc,thus, it can effectively reduces other unrelated boundary segments and correctly extracts arc. Arc is able to provide more accurate data for the follow-up ellipse detection.(3) Because arc extracted above is generally part boundary of ellipse, so in order to obtain complete edge points for each ellipse, this paper has designed arc grouping algorithm. This method considers the connections of both arc and arc ends to determine whether they belong to the same ellipse, so it achieves correct grouping of arcs. Comparing with arc. arc group of the same ellipse can provide more complete point data for ellipse detection.(4) Using numerically stable direct least square ellipse fitting algorithm as the main method of ellipse detection algorithm, this paper has designed ellipse detection based on arc group. At the technical aspects of algorithm implementation, this paper uses Java and MATLAB mix programming, given full play to the rich dynamic data types in Java and the advantages of matrix operations, eigenvalue and eigenvector calculation in MATLAB,which makes the implementation process of algorithm more flexible and reduces the complexity of program development.Finally, with the designed ellipse detection algorithm based on are group, experiments have carried out on the synthetic images and real images. It has accurately and effectively accomplished ellipse detection and parameters extraction. and results show that the algorithm is effective and feasible.