Research On Color Image Segmentation Technology Based On Multiple Spaces

Digital image processing is a method and technology that uses computers to remove noise,enhance,restore,segment,and extract features from images.Image segmentation is a fundamental and important task in image processing,and has been widely used in many fields such as medical imaging,face recognition,and so on.The purpose of image segmentation is to reasonably segment an image into disjoint regions.Color image segmentation is a major difficulty in the field of image segmentation.In the past few decades,many color image segmentation methods have been proposed,and the existing image segmentation methods are mainly divided into the following categories:region-based segmentation methods,edge-based segmentation methods,special theory-based segmentation methods,and threshold-based segmentation methods.Due to the fact that many image segmentation models are non-convex and non-smooth,and it is difficult to solve them,this thesis focuses on the research of variational models for non-convex regular image segmentation for color image segmentation tasks.The main research content and innovation points are as follows:1.In this thesis,we propose a three-stage image segmentation model based on L₁-αL₂regularization and MS model.This model uses non-convex L₁-αL₂ regularization as a priori information of the image,and uses non-convex approximation to the non-convex Hausdorff measure in the MS model,in order to extracting more boundary information.In the first stage,the convex difference algorithm（DCA）and alternating direction multiplier method（ADMM）are used to jointly solve the model to obtain the image smoothing results.In the second stage,color space is converted using a strategy of combining RGB space and Lab space.In the third stage,the smooth image obtained in the first stage is segmented using the threshold value determined by the K-means clustering method.Finally,numerical experiments verify the effectiveness of the proposed method.2.In order to better solve the difficulty of color image segmentation tasks,we propose a segmentation model based on quaternion and L₁/L₂ regularization for color image segmentation tasks.In this work,based on L₁/L₂ regularization,we propose a two-stage strategy for color image segmentation,using the semi-proximal alternating direction method of multipliers（s PADMM）to solve the proposed model,and considering the inherent color structures in different channels,we use a quaternion matrix to represent color images.In the experimental part,we conduct parameter analysis,synthetic image segmentation,and natural image segmentation.We also add noise and blur to some test images,and then used our new model to restore and segment degraded images.The experimental results show that the proposed method has certain advantages.3.In the second work,we propose a color image segmentation model based on L₁/L₂regularization.Due to its non-convex regularity,we prove the existence of solutions to the model proposed in the second work,and also proved the convergence of the algorithm used.By introducing different non-convex regularizations in different spaces to segment color images,both visual effects and numerical experiments have shown that the methods proposed in this thesis have advantages over existing color image segmentation methods.