Image Recognition Based On Geometrical Shape

In the identification and understanding of human's visual perception, shape is an important parameter of expressing the information of objects. Extracting each shape from the image is a step which need be carried out for most target recognition algorithms.Corners are important local features of an image, and are the feature points that can be fully describe the shape of the object. In the field of the image registration based on the characteristics, image understanding and pattern recognition and so on, the corner extraction is of great significance. Linear, free-form curve is a common geometric shape, also is other more complex geometrical element. So the extraction of line and free curves is of important significance in the practical application. The paper was studied in the following areas:①Image Segmentation is an important issue of image processing and machine vision. In the image collection, distributing of pixels sometimes exists cross phenomena and some of the object edge has some burr. A partial density threshold segmentation method and a segmentation of removing the burr are proposed in the paper. These methods resolve the above problems.②In the field of image processing and pattern recognition, chain code is a very common coding technology used in the expression of lines, planar curves and a border region. Corner detection plays an important role in pattern recognition and machine vision. The paper proposes a new corner extraction algorithm based on average chain code. Curvature calculation in the common corner extraction algorithm based on Freeman chain code will be converted into a margin of average chain code, and the best threshold will be found. Through the control of the threshold, it has accurate extraction of border corners. Another, a simple and efficient algorithm of the line approximation of free curves based on Freeman chain code is also proposed. This method not only applies to linear, circular and non-circular curve, but also applies to free curve which have complex shape and can not be described by primary analytic function.③Aimed at the limitations of clustering of the traditional calculation of similarity based distance, a method of point-set clustering based topology geometry-figure is proposed in the text. The method can extract some kinds of objects with the geometric characteristics wishing to be needed from the discrete point-sets. And, extracting a topology geometry point set from a discrete point set is a problem which Hough transform(including the promotion and improvement of the Hough Transform), the chain code, such as traditional methods in the field of image recognition technology can not be achieved. The experiment shows that the method can detect effectively the curve points of small curvature, to some extent,it overcomes the limitation of the method based distance. It can be applied in the areas of engineering drawings recognition,computer vision,remote sensing recognition and so on.The results are developed in the VC++ environment, and they show that, a series of ideas and algorithms have the feasibility and correctness.