Research On Texture Classification Method Based On Complex Network Model

Texture is an extremely important visual clue,which provides a reliable basis for people to recognize and analyze things.As the research boom in fields such as artificial intelligence and computer vision continues to heat up in recent years,texture analysis has become a hot field of basic research.Among them,static texture classification and dynamic texture classification,as the main analysis methods,continue to play an important application value in many fields such as medical diagnosis,fire warning,smart transportation,industrial production,and satellite detection.The complex network model mainly includes two parts: First,use the texture-to-network mapping relationship and the characteristics of the texture itself to construct a complex network model to describe the dynamic texture and the static texture respectively.Achieve the multi-scale analysis of texture through the dynamic evolution process of the complex network.Secondly,it realizes the feature extraction of the network(texture)combined with the related metric describing the topological structure of the complex network.Based on the above theoretical foundation,this paper mainly studies the classification method of static texture and dynamic texture based on complex network model.The main work of this paper is as follows:1 A static texture classification method based on complex network model and adaptive threshold evolution algorithm is proposed.Use the samples to be classified to build a Gaussian Pyramid model before complex network modeling.The initial rule network is constructed through the relationship between pixels and their neighborhoods in different levels of the texture of the model,and the initial rule network is dynamically evolved using the proposed adaptive threshold evolution algorithm(ATEA).On the one hand,it generates complex networks with different topological structures corresponding to different types of textures,and on the other hand,it realizes multi-scale analysis of texture samples.Next,the related measures describing the complex network topology are used to characterize the complex network,so as to realize the feature extraction of the texture.This paper attempts to apply two important measures of complex network topology,namely,Network Centrality Measure(GC)and Node Degree Entropy(NDE).On the basis of the extracted features,the nearest neighbor classifier(NN)and SVM is used for classification,and the correct classification rate(CCR)is used as the evaluation index.Through comparison with other algorithms,the classification performance of the proposed method is analyzed.2 Combining the characteristics of dynamic texture,the static texture classification method based on complex network model is extended to the dynamic texture classification method.the dynamic texture is modeled as a three-dimensional complex network,and combined with the adaptive threshold evolution algorithm(ATEA)to achieve the multi-scale analysis of the dynamic texture.Before feature extraction,the network is divided into time network and space network according to the time axis to describe the "temporal characteristics" and "spatial characteristics" of the dynamic texture respectively.Combining complex network correlation metric for feature extraction,the nearest neighbor classifier(NN)and SVM is used to achieve dynamic texture classification,and the correct classification rate(CCR)is compared with other methods to analyze the classification performance of the proposed method.