Research On Image Segmentation Based On General Fuzzy Operation And Active Contour Model

Image segmentation is a very important problem in image processing, pre-processing of images can only obtain optimal approximation of original image, so image segmentation is necessary before feature extraction and recognition. The main evaluation criteria for image segmentation algorithm is effection and efficiency, whether image segmentation is accurate or not has a direct influence on image analysis results. The segmentation efficiency is a important condition for the application of computer vision system. To improve the segmentation effection and efficiency, we fulfill the following work:There is an obvious difference between General Fuzzy Operator(GFO) and other edge detection which based on edge magnitude. But some key parameters of GFO are manually tuned by designers, which has influenced the detection effect, this limits the practical application. To solve this problem, a new method which applies genetic algorithm and Otsu theory to the determination of GFO parameters is proposed. This method can obtain the parameters self adaptively. Then the improved GFO algorithm is applied to the complex image edge detection which contains a great amount of pixels and gray values, it greatly improves the effect of image edge detection and makes it easier to be used in image segmentation.Parametric Active Contour Model brings feedback that high processing extracts low contour through a energy function which can reflect object contour and gray, it is a popular model which used for target contour extraction in image processing recently. The classical model has shortages, such as being sensitive to original contour and the computing time being rather long, converging concave difficultly, et al. An improved model is proposed, where the improvement are in energy function and numerical iteration method. It shows better performance on convergence speed and more accurate than the original one, thus overcomes the shortages of the classical model, such as small search region and converging concave difficultly, the computing time being rather long et al.Geometry Active Contour Model overcomes the shortages of Parametric Active Contour Model that can not treat with the change of topology naturally, and solves theproblem that Parametric Active Contour Model can not complete. Level set method promotes the development of Geometry Active Contour Model that has attracted increasing research interests recently. In order to overcome the shortages of level set, such as it needs an level set function updating for all the points which are defined in the whole image region, complex computation, time consuming et al. An solution combined with PSO,spline interpolation algorithm and narrow band of level set is presented, it greatly reduces the calculation time and avoids noise affection, improves segmentation results.C-V algorithm of Mumford-Shah model is adapt for the contour extraction that gradient is reasonable or not in image. This solves the case of much fuzzy target boundaries, and prevents the"boundary leaking"problem. Then it needs update for all the points which are defined in the whole image region in each iterative process, consume large time, and it is unable to make use of narrow band directly. A solution is proposed, C-V algorithm of M-S model can combine with narrow band effectively, and it can greatly reduce computation and improve efficiency of image segmentation. In addition, a improvement in C-V algorithm is proposed, an iteration algorithm and corresponding stopping criterion for iteration is presented. It can solve the problem of large computation in C-V algorithm.For proved the effect of the proposed algorithm and the application theory to practice, an robot vision tracking platform based on MOTOMAN-UP6 robot is designed. In the process of design and achievement of the robot vision tracking system, target image recognition technology which is unavoidable in computer vision field is discussed, a recognition method based on RBF neural networks based on Hough transform is proposed ,and the details of the traking predicition is presented. The robot vision tracking experiments show the proposed image segmentation algorithm has good performance and prove that the image tracking software designed in this study is effective and accurate.