Research On Multivariate Image Analysis Methods For Texture Colour Feature Images

Image analysis is a technique for understanding,interpreting,and classifying images using mathematical tools.With the development and diffusion of computer technology,image analysis techniques have also become increasingly important because of their ability to perform fast,non-invasive and low-cost analysis of targets.While classical image analysis techniques are mainly aimed at grey-scale images,the popularity of multispectral images has led to an increased focus on the analysis of data from multiple channels of an image.In the 1980 s,the field of chemometrics introduced a powerful image analysis tool that could analyse data on multiple variables in multivariate images,providing new insights into the spatial and spectral information of such images,a tool known as multivariate image analysis techniques.Multivariate image analysis techniques apply to many images from grey-scale to hyperspectral images and are progressively being used in areas such as microscopic imaging,industrial control,and mineral processing.As the application of multivariate image analysis techniques spreads,their lack of analysis of image texture features is becoming a pressing problem to solve.In this thesis,the solution to the problem is given after an exhaustive theoretical derivation and experimental analysis,based on the basic principles and mechanisms of multivariate image analysis techniques and the nature of image texture features,and the application method is proposed based on this solution.The main points of this article are as follows:(1)The background and significance related to the research of this thesis are described,the research on multivariate image analysis and image texture features are reviewed,and the multivariate image analysis techniques are compared with traditional image analysis techniques.A detailed description of the object of study,the qualities and principles of the mathematical tools and application methods of the multivariate image analysis technique,its analysis,experiments and the mechanism of the lack of analysis of texture feature information by the multivariate image analysis technique are given,leading to the main research content of this thesis.(2)In response to the problem that multivariate image analysis techniques lose the spatial connection between pixels in the multivariate image unfolding stage,the solution of embedding the local texture information of the region where a single pixel in the image is located in the pixel intensity is proposed,and implementing the solution is completed by combining the greyscale co-occurrence matrix and its statistics and the sliding window method to give a method for calculating the texture feature image of a two-dimensional image.The texture feature image is got for each channel of the RGB image and added to the original channel image to get the texture-colour feature image.The multivariate image is constructed by superimposing texturecolour feature images,and based on this,a multivariate image analysis method for texturecolour feature images is proposed.(3)Two image segmentation methods are proposed based on texture-colour feature image multivariate image analysis theory,which integrates the texture and colour features of images.One is a multi-threshold segmentation method for texture colour feature image multivariate image analysis,which first selects the score image that reflects the features of the region of interest from the score image got by multiplex principal component analysis,then uses threshold to segment the part containing the region of interest from the selected score image,and finally takes the intersection of all segmentation results to complete the segmentation.The second is a decision tree and pixel-level classification based image segmentation method,which uses the segmentation results of the region of interest and the score matrix to construct a decision tree based on the C4.5 algorithm,using the decision tree to complete the pixel-level classification of the image.Experiments are conducted on the image dataset to verify the effectiveness of the segmentation method.