Research On Analysis And Classification Of Image From The Perspective Of Dynamic Network Analysis

With the rapid development of computer and information technology,image representation and feature extraction,as a key part of image target recognition,play a vital role in machine vision,pattern recognition and other fields as well as image analysis and recognition.Texture and shape features are also important features of image representation research.In recent years,among the many important frontier research fields of computer science,network science and its applications have developed into one of the important and emerging frontier research fields.The rapid integration of multiple disciplines and novel research perspectives has attracted many outstanding features.Many scholars are actively involved in exploratory research on the theory and its applications.This paper mainly studies the representation of image texture features and shape features from the perspective of dynamic complex network analysis,and applies them to the classification of texture images and shape images.Shape and texture are extremely important features in human communication,and are one of the most important visual attributes to represent objects in pattern recognition and image analysis.They provide the most relevant information about objects for identification and classification tasks,and play an important role in object recognition,object detection,shape retrieval,texture recognition and other applications.Using dynamic complex network to analyze image shape and texture has a very wide application prospect in the field of image processing and analysis.The main research work of this paper is as follows:1)Feature extraction and pattern classification are important issues in pattern recognition.Among them,texture analysis and texture feature representation are one of the most basic problems in the field of graphic image analysis;at the same time,it is also a research hotspot in machine vision and other fields.Based on the novel classification method proposed by Backes Andre Ricardo et al.of "constructing accompanying dynamic evolutionary complex network" and generating "high-dimensional interpretation vector for contour image classification",this article first studies the classification of texture image libraries.This paper suggests that during data preprocessing,the data set image samples should be segmented twice to minimize the amount of calculation of the algorithm;this paper also improved the "network vertex randomization,sampling acquisition" strategy when constructing a texture network.Under the appropriate threshold conditions,the calculation amount of the algorithm can be further reduced.The article introduces in detail the generation strategy of "high-dimensional feature interpretation vector" for texture image data classification,and the extraction process of texture image "geometric digital features".The texture image fingerprint vector constructed based on the improved process can effectively extract the inherent digital features of the texture image.Numerical results show the effectiveness of the algorithm and the robustness of such methods to rotation operations and noise interference.This type of method has certain industrial application prospects.2)Secondly,after successfully applying this method to the study of texture image classification,this article further extends the method to shape analysis and classification research.This paper tries to use a variety of edge operator contour extraction methods,and compares and analyzes the influence of different edge operator extraction algorithms on classification algorithms based on complex network analysis.In graphics images,the accuracy of edge contour extraction determines its physical characteristics.Whether useful and undisturbed contours affect subsequent modeling and classification is one of the topics that need to be discussed in this paper.In this paper,common first-order and second-order edge operators and rough boundaries are used to extract the contours of the shape image data set for modeling and classification.By comparing the results of numerical experiments,the selection criteria of image edge contour extraction methods in actual application scenarios are preliminarily discussed.Build problems.In view of the fact that the limited static statistics are difficult to reflect the inherent characteristics of the generated companion network,when extracting complex network features,this paper considers the interaction between dynamic changes and static structure,and obtains higher-dimensional feature vectors from the perspective of evolutionary networks Combine it as an explanatory variable for further classification.By applying the algorithm to Etu10 Silhouette database and Kimia216 database for numerical experiment verification and analysis,some basic application selection criteria for "contour extraction edge operator" and "reasonable classification method" are found.At the same time,the high-dimensional geometric signature feature extraction process of the shape given in this paper has certain versatility and has a certain potential application value.3)Finally,this paper carried out the research on the influence of boundary extraction operator on the face classification algorithm based on complex network,and gave the specific effect analysis of another application scenario of the image classification algorithm based on complex network developed in this article.By constructing a complex network with accompanying dynamic evolution boundary and generating high-dimensional explanatory variables,a classification analysis method has achieved a good classification and recognition effect in processing human faces;Furthermore,in order to eliminate the influence of common first-order and second-order contour extraction algorithms on the classification accuracy,this paper also uses the gray threshold evolution method to extract contours for experimental verification,and the numerical results have been significantly improved.This shows that the complex network analysis method that constructs the complicated network with the accompanying dynamic evolution boundary to generate explanatory variables is also feasible in face recognition.