Research On Rapid Extraction Of Main Structure Of Texture Pattern

The image can convey a wealth of information simply and intuitively,and it has irreplaceable importance as the most commonly used information carrier in modern society.The use of image processing technology to extract information from the image is very helpful for understanding the image and using the image,especially in the context of the highly informatized society and the explosion of the number of image data.The image often contains rich texture information.The diversity of the shape and characteristics of the texture interferes with the extraction of the image information.The rapid extraction of the main structure of the texture image can be used as an essential part of image processing techniques such as image edge detection,image segmentation,and tone mapping.Machine vision applications such as target recognition and motion analysis also require rapid extraction of texture master structures to reduce the interference of texture information.According to the principle of the weighted least squares model in the optimization model method,a penalty operator has been proposed,which can generate different weights in the texture area and the main structure area,thus,the two regions are smoothed to different degrees to achieve the purpose of eliminating the texture and getting the the main structure.The texture region image sample library and the main structure region image sample library were established to verify the validity of the weight operator.Tried to think about the problem of determining the main structure area from the perspective of deep learning and explored the low-dimensional feature representation of high-dimensional texture image matrix data.On the basis of the Tensorflow deep learning framework,constructed and trained the neural network to determine the main structure area and decide the right value.Through examples,the difference in terms of effect and time consumption between the way to extraction of main structure of texture pattern by using weight operator directly to solve the weighted least squares model or using the neural network to determine the main structure area has been determined.The weighted least squares model involves the solution of large sparse matrix equations.In this paper,a layered preprocessing method has been used to process the coefficient matrix,and the condition number of the sparse matrix the equation can be reduced without affecting the extraction effect of the main structure of the texture image,thereby reducing the time consumption of solving the matrix equation.A library of texture image samples with different pixel sizes was established,and a sample library was used to verify the acceleration effect of the preprocessing method in this paper on the solution of large sparse matrix equations in the extraction process of the main texture image.