A New Method Of Gray Image Recognition Based On Complex Network

Image recognition technology is an important part of computer vision.It is widely used in surveillance video,traffic surveillance,human-computer interaction,mobile phone recognition and unlocking,license plate recognition and so on.There are many methods of image recognition at present.According to the different recognition features used in the process of recognition,it can be roughly divided into four parts: region-based algorithm,model-based algorithm,contour-based algorithm,feature-based algorithm.According to the different algorithms used in the recognition process,it can be roughly divided into three parts: image recognition method based on filtering theory,image recognition method based on mean shift,and image recognition based on partial differential equation.However,because these algorithms are inextricably linked to the positions and orders of the pixels in the image,the recognition accuracy of the image is affected when it is rotated,translated,and scaled.Therefore,exploring and discussing the effective improvement of existing image recognition methods has engineering practice significance.In the past ten years,the application research based on complex network with graph model is the important research direction which received great attention in the field of engineering.It is widely used not only in the fields of network communication and control engineering,but also in the field of image recognition application.The complex network is composed of nodes and links,and regarded as the topology network.The structure of the network is only related to the interconnection relationships between nodes,regardless of the orders and locations of nodes.Therefore,theoretically,the image recognition by using complex network may reduce the impact of images on the accuracy of recognition when rotation,translation and scaling occur.It is worth noting that the existing image recognition methods based on complex network have mainly the following two shortcomings:(1)In the complex network modeling,all the pixels of the gray image are chosen as the nodes of the network,that is,the number of nodes in the associated network model is the same as the pixels in the original image,which results in the quite slow calculation of the network.(2)In complex network modeling,the weighted network is always chosen as a fully weighted graph,which means that there are invalid feature parameter calculations(in time and storage).This not only drags down the calculation speed,but also affects the recognition accuracy.For the above problems,this paper firstly uses the watershed algorithm to segment the gray image,and divides the image into multiple connected regions.Based on the principle of regional pixel similarity,the representative pixels of the extracted regions are used as nodes,by which it may greatly reduce the number of nodes in the image complex network model,and improve the calculation speed of image recognition.In view of this,this thesis focuses mainly on constructing the network connection relationship by using Euclidean distance and absolute gray scale difference,respectively.Meanwhile,the selection method of network topology feature parameters is improved.Finally,by using the interest points in the image,the accuracy and speed of image recognition have been improved.Compared with the some existing image recognition algorithms,the results of numerical simulation shows the main advantage proposed in this paper.The main advantage is that the number of nodes in network modeling is greatly reduced,and thus the complex network is more sparse,which results in the faster calculation to the image recognition algorithm.