Research On Structure-Balance Complex Network Based On Novel Image Recognization Methods

Image recognition is one of the important means to realize intelligent technology,which is widely used in finance,transportation,artificial intelligence and other fields.Image feature extraction plays a vital role in the recognition effect.Scholars have proposed many methods for feature extraction.There are include several image recognition methods,for example,image recognition based on geometric feature,image recognition based on statistical feature,image recognition based on artificial feature design,image recognition based on convolutional neural network and so on.In these methods,the accuracy of the algorithm will be affected when the image is rotated or translated.Therefore,it is of significant practical engineering significance to propose a method that does not depend on the position and order of pixels.In the past decade,complex network methods have achieved a lot of research work in the field of image recognition.The complex network constructs a model based on graph theory,abstracts pixels as network nodes and extracts the topology measurement parameters of the network as the features of image recognition.Some image recognition methods based on complex network show that using pixel points as nodes to build a complex network model for grayscale images,the model's topological parameters do not depend on the order and position of the nodes.However,large scale of network nodes,difficult modeling and high network complexity are the disadvantage of image recognition methods based on complex network,which make long operation time for extracting topological feature parameters.Therefore,on the basis of previous work,this paper proposes a complex network image recognition method based on structural balance.In this method,pixels are used as nodes,and the initial network is generated by constructing the connection relationship between the nodes by the gray product,and then the initial network is mapped by the Hadamard transformation template to establish a structural balance network of the grayscale image,and the image feature parameters are extracted based,this main work of this article is as follows:1)This paper proposes to use the gray value product between two nodes as the connection edge weight.Compared with the use of Euclidean distance to generate connected edges of the network,this modeling method in this article not only reduces the amount of calculation in the modeling process and reduces the difficulty of modeling,but also reduces the amount of calculation for image feature extraction.2)This paper proposes to use the Hadamard transformation template to generate the structural balance network of grayscale image,and establish the auxiliary network of the gray-scale image by using the selection function to assist in extracting the characteristic parameters of the image.Compared with the traditional method of building a grayscale image's complex network model based on distance thresholds,the method in this paper reduces the number of threshold function selection,thus increasing the calculation speed partly.3)In this paper,the feature extract from the structure-balanced network be used to image recognition feature parameters.These feature parameters have network topology meaning,thus reduce the dependence of traditional image recognition in complex network on topology parameters.This paper effectiveness of this paper is verified Leaf database,HPI database,Yale database,and this method is compared with the image recognition method of traditional complex network.Finally,the result show the proposed approach has better effects than the rivals' methods in modeling time of and the computation time of feature parameter extraction.