A New Image Identification Method Based On Contour Shape And Complex Network

Target image recognition is an important part of machine vision technology,the accuracy and speed are two important measure indexes of the target image recognition algorithm. At present, there are many methods of image recognition, the most widely used and most effective method is the image recognition which is based on shape. Compared with other methods, the identification method based on contour can not only simplify the recognition process, reduce the storage space in the process of computer processing, but also has a high recognition speed and accuracy.However, due to the order of shape contour points, the position and arrangement order of the contour points have great influence on the contour recognition, especially when the image contour is rotated, translated and scaled in the plane, motion invariance has become an important consideration for the classification of image shape contour recognition methods. Therefore, it is very necessary to establish a quantitative method to analyze the boundary shape of the target image with little or no dependence on correlation and sequence.As a new research field, complex network has been widely used in the fields of life science, engineering science and social science. Complex network is a kind of network model which is described by graph theory to describe complex system; this model is composed of a large number of nodes and the connections between nodes.It is noting that the relationship between nodes and the sequence of the topological characteristics, based on the Euclidean distance, is not independent. The existing research results show that it is an effective way to solve the problem of image boundary shape recognition without dependence and sequence by using the topological characteristics of complex network to improve the quantitative method of image contour recognition.It is worth noting that Euclidean distance is the basis of existing image contour network modeling method, and Euclidean distance in the calculation of the distance between the nodes of the network, ignore the difference between horizontal and vertical coordinate flaw, easy to generate redundant node connection edges, and increase the complexity of the network. In view of this, this paper based on the existing research, works mainly in the "shape contour recognition method combined with complex network method", "new image contour extraction method" and "improvement the measure distance method between network nodes " and so on. This paper proposes a new method to measure the basis of network modeling with spherical distance, and found out the following result:the method based on Euclidean distance are more suitable for processing the image with linear boundary, the method based on spherical distance is more suitable for the processing of circular arcs. The method in this paper, in a certain extent, gives the solution to the problem that the network model based on Euclidean distance does not adapt to the processing of circular arc image. These results in this paper further enrich the theoretical approach to reveal the characteristics of the target image by using the topological parameters of complex network model.Compared to the traditional Euclidean distance modeling method, the methods in this paper have the following advantages:(1) Under the same threshold, the number of nodes in the network is smaller, and the storage space occupied by the computer processing is smaller, and the computation speed is more fast; (2) For the network model established by the method of spherical distance, the degree distribution is more in line with the power-law distribution. This means that the topological characteristics of complex network mode proposed in this paper are shown more apparently, which can effectively adapt to changes in the image edge contour, and contribute to improve the recognition rate and recognition speed.