Research On Sub Pixel Edge Detection Technology

Edge detection of the digital image is very important basis in image analysis fieldsof image segmentation, target recognition, region shape extraction, etc. In imageunderstanding and analyzing, the first step is always edge detection. At present, theedge detection has become one of the most active subjects in machine vision researchfields, which has very important theory significance and practical application value. Thedetection accuracy of the traditional edge detection algorithms can only reach a pixellevel. However, with the rapid development of science and technology, industrialdetection and other applications for accuracy requirements is increasing. The traditionalpixel level edge detection algorithms have been unable to satisfy the need of practicalmeasurement, so we need the edge detection algorithms with higher accuracy, namelythe sub pixel edge detection algorithm. In this paper, the sub pixel edge detectiontechnology is studied deeply, and the research contents include the following severalaspects:（1）In this thesis, the research achievements and the present developmentsituation of the digital image edge detection technology and the sub pixel edge detectiontechnology at home and abroad is described. And the research significance of the subpixel edge detection technology is discussed. At the same time, the future developmenttrend of the sub pixel edge detection technology in image processing field is analyzed.（2）Firstly, the correlation contents of the digital image are introduced. Thenseveral classic pixel level edge detection algorithms are studied, which mainly includeRoberts operator, Sobel operator, Prewitt operator, Laplace operator and Canny operator,etc. And these several algorithms are simulated with the Matlab programming tools.Their advantages, disadvantages and applicable scope have been compared. At last, thethesis analyzes the key factors which influence the quality of edge detection.（3）Several sub pixel edge detection algorithms which are commonly used are studied, including the space moment algorithm, the gray moment algorithm, the Zernikemoment algorithm and the digital correlation algorithm, etc. Meanwhile, they aresimulated with Matlab programming tools. Then, two different target images namelystandard hexagonal images and nut image are simulated respectively. Then according tothe experimental results, the performance of the various algorithms in accuracy,detection speed, anti-noise ability and other aspects have been compared.（4）According to the detection accuracy of the traditional pixel level edgedetection algorithms is low, and the sub pixel edge detection algorithms have improvedthe accuracy, but the detection speed is slower. Therefore, a spline interpolation subpixel edge detection algorithm based on improved morphological gradient is proposedin the thesis, which is combined the improved morphological gradient filter operatorwith the cubic spline interpolation algorithm. Then, this improved algorithm issimulated with Matlab programming tools, which is compared with the traditional pixellevel edge detection algorithms in the thesis. And the experimental results show that theimproved algorithm can detect the edge of digital images accurately, and the detectionaccuracy is higher than the traditional edge detection algorithm obviously. Meanwhile,the detection time of the improved algorithm has been compared with the space momentalgorithm, the gray moment algorithm, the Zernike moment algorithm and the digitalcorrelation algorithm, the results show that the improved algorithm not only can ensurethe accuracy, but also reduce the detection time.（5）First of all, the polynomial fitting algorithm, the ellipse fitting algorithm, thegauss surface fitting algorithm and the sigmoid curve fitting algorithm are studied. Theiradvantages and disadvantages are compared. Compared with the quadratic curve fittingalgorithm, the calculation process of them are complicated relatively. So in this thesis,the quadratic curve fitting algorithm is studied intensively and improved. An improvedsub pixel edge detection algorithm based on quadratic curve fitting is proposed in thethesis. When the traditional sub pixel edge detection algorithm based on quadratic curvefitting is used for detecting edge, the adjacent pixels of edge points are selectedrandomly in its eight neighborhoods, which are used in edge extraction formula. But the proposed algorithm selects certain directional points in the adjacent pixels (theexperimental results show that when we select by135°direction, the effect is best),then the image edge are determined through the sub pixel edge extraction formula basedon curve fitting. Finally, the improved algorithm has been implemented with the Matlabprogramming tools. The experimental results show that the proposed improved subpixel edge detection algorithm based on curve fitting can detect the edge of the digitalimages accurately, and its detection effect is superior to the original algorithm based onquadratic curve fitting.