Image Segmentation Based On Indiscernible Relation And Image Registration Based On Contour

Image segmentation and image registration techniques have a wide range of applications in many research fields, they are also hot spots at domestic and overseas. Image segmentation is the premise of completing image visual analysis and pattern recognition, only to complete accuracy segmentation, it can be possible to realize high-level image analysis and image understanging. Image registration is both basic of image target recognition and3-d reconstruction, as well as the key of image fusion technology, registration results will directly affect the quality of the subsequent fusion.It is studied the image segmentation and image registration techniques in this paper, and also do some new try for each technique:research on rough set model of granular computing theoty and the application of rough set in image segmentation, then it is proposed a image segmentation mehod based on indiscernible relation; research on image registration methods, then it is proposed a method of extracting contour with the new definition’s normalized neighborhood variance, and on this basis to find out the registration parameters to reduce computation cost, realize registration.The main work of this paper are as follows:1. It is proposed an image segmentation method based on rough set theoty: this method first uses classification thought of rough set, chooses gray value and normalized neighborhood variance as condition attributes, then take the informations which an image expresses as a knowledge expression system, divides an image into different equivalence classes based on indiscernible relation, finally take these classes polymerize to form the segmentation image. Due to the selection of threshold is a key of synthesising classes, so it is also proposed how to obtain the better thresholds. This method takes into the impact of image noise, use mind evolutionary algorithm(MEA) to optimize the threshold parameters, to avoid into local minimum at the same time improving operation speed.2. For segmentation, it also introduces some representational image segmentation methods, then uses these methods to compare with the proposed method, in order to verify the effectiveness of the proposed mmethod. The experiments results show that for normal images, the segmentation images have better uniformity and denoise effects, the edge of images is smooth and clear; for medical images, this method can better extract the lung tissue, and bronchi fracture less, so the proposed method is an effective method, it has better stability and convergence speed, also has application value in practice.3. It takes human brain medical images as the research objects, takes head as a rigid, uses rigid has contour invariance, then it is proposed a image registration method based on contour:this method extract contour with the new definition of normalized neighborhood variance, then uses the classical mechanics of torque spindle to respectively calculate the centroid and the angle of spindle and axis, obtain parameters of translation transformation and rotation transformation, realize aligment of registration images’centroid and spindle, complete the registration process.4. For image registration, it uses experimental results which takes classic operators of edge detection to extract target contour then registration to compare with the experimental results that is obtained by the proposed method, then verify the feasibility of the proposed method. The experimental results show that the proposed method achieve fast, can meet the requirements of prophase image registration.