Research On Color Texture Classification Method Based On Signal Processing On Graph

The data structure of graph can use vertices to represent samples,and edge weights to represent the relationship between samples to realize the modeling of irregular data.It has been widely used in many fields such as neuroscience,image analysis,swarm intelligence and neural network.Therefore,it is crucial to integrate the classical signal processing framework with graph theory and study the processing tools in the signal processing on graphs framework.Compared with real numbers and complex numbers,quaternions have more phase information.In this paper,the graph wavelet filter bank is extended,and the construction method of quaternion graph wavelet filter bank is proposed.Color and texture are important visual information for people to perceive the world.With the development of multimedia technology,image classification using color and texture information is also an important research direction.Quaternion graph wavelet transform can combine the structure and relationship between pixels,provide richer information at different scales,and can analyze color image texture more effectively.This paper mainly studies the construction method of quaternion wavelet filter bank on graphs and the color texture classification method based on the quaternion wavelet transform on graphs.The main work is as follows:(1)Aiming at the limitation that the coefficients obtained by real wavelet transform on graphs and complex wavelet transform on graphs lack part of phase and cannot process interrelated multi-channel signals at the same time,the wavelet filter bank on graphs is extended,and a construction method of quaternion wavelet filter bank on graphs based on quaternion Laplacian matrix is proposed.According to the design principle of the two-channel wavelet filter bank on the bipartite graph,the quaternion Laplacian matrix satisfying the non-negative real eigenvalues is constructed and proved.Quaternion Fourier transform on graphs and filtering on graphs are defined based on eigenvalue decomposition of quaternion Laplace matrix.The phenomenon of spectral folding of bipartite graphs of quaternion values is demonstrated.The quaternion wavelet filter bank on graphs is constructed and proved to satisfy perfect reconstruction and orthogonal decomposition.The effectiveness of this method is proved by experiments.(2)Aiming at the problem of how to make full use of color and texture features to make the classification of texture images more effective,using the above quaternion wavelet filter bank on graphs,a color image texture classification method based on quaternion wavelet transform on graphs is proposed.Perform quaternion wavelet transform on graphs on the color image texture,perform local quaternion singular value decomposition on the generated coefficients to extract the main information,perform Weibull modeling on the effective singular value vector distribution obtained from the decomposition,and realize the feature extraction from color texture.The color texture is classified by the nearest neighbor classifier measured by KL divergence,and the effectiveness of this method is verified through classification experiments on two color image texture datasets.