The Research On Adaptive Morphology Theory And Its Application For Watershed Image Segmentation

The mathematical morphology theory mainly utilizes structural elements similar to"probes"to obtain the geometric structure of the analyzed image,and its basic ideas and methods significantly impact image processing technology.The basic ideas and methods have significantly impacted image processing techniques and have been used in applications such as machine vision,industrial inspection,biomedical engineering,aerospace remote sensing,military science,and railway safety monitoring.The classical mathematical morphology theory uses structural elements of fixed shape and size to perform morphological operations on images.Some problems cause changes in the information of image target features,such as generating new artificial targets,changing the edge positions,and losing small targets.To address those problems,the thesis focuses on the analysis and research of how to construct adaptive structural elements with variable shape and size according to the local features of the image,how to establish the basic morphological operations corresponding to the adaptive structural elements,and apply the constructed adaptive structural elements to the watershed image segmentation.The main research work and innovative results of the thesis include the following:（1）The research of adaptive structure element construction method based on the vector field of neighborhood gray level difference change and relative densityIn order to solve the problem of shifting the edge contour of the image target in the morphological operation by traditional structural elements,the adaptive structural elements are constructed by combining the neighborhood gray level difference change vector field and relative density.First,the image is smoothed using the neighborhood gray difference change vector field to improve the separability between the image target’s internal and external pixel points.Then,a boundary degree function based on the relative density is defined to determine the strong boundary points in the local area of the image,and the adaptive structure element is composed of the strong boundary points in the local area of the image.The experimental analysis shows that the Abdou-pratt quality factor,peak signal-to-noise ratio,and structural similarity of this class of adaptive structure element are significant,and the deviation of mean square and gradient magnitude similarity is slight as morphological operations are performed on the image,which indicates that it has good ability to suppress the offset of image target edges and maintain the practical details of the image.In addition,using it to perform erosion and expansion operations on images with relatively few detailed features and relatively regular target geometry,the mean square deviation is relatively small.The structural similarity is relatively significant,indicating that it is more suitable for processing such images.（2）The research of adaptive structure element construction method based on the local density and symmetry of pixel point neighborhoodIn order to solve the problem that the loss of image target edge information occurs by conventional structural elements for morphological operations,the mathematical principles of neighborhood local density and symmetry are applied to the construction of adaptive structural elements.First,the image is smoothed using the vector field of neighborhood gray difference change.Then,the local density function,symmetry function,and coefficient of variation of the neighborhood of the image pixel points are defined to determine whether the center pixel point of the local area is a boundary point and whether the adaptive structure element is composed of the boundary points in the determined local area of the image.The experimental analysis shows that the Abdou-pratt quality factor,peak signal-to-noise ratio,and structural similarity of this type of adaptive structure element are significant,and the deviation of mean square and gradient magnitude similarity is slight when morphological operations are performed on the image,which indicates that it has good ability to keep the target edge information and other helpful information.In addition,using it to perform erosion and expansion operations on images with relatively more detailed features and relatively irregular target geometric structure,the peak signal-to-noise ratio is relatively small,and the deviation of gradient magnitude similarity is relatively large,indicating that it is more suitable for the processing of such images.（3）The research of adaptive structure element construction method based on fuzzy C-means clusteringIn order to solve the problem of reduced accuracy of image target edge localization arising from morphological operations performed by traditional structural elements,the idea of fuzzy C-means clustering is introduced into the construction of adaptive structural elements.First,the optimized fuzzy C-means clustering objective function clusters the image local regions.Then,the edge characteristics of classes and classes after clustering are used as the primary basis for constructing the adaptive structure element,which consists of the boundary points between classes in the image’s local area.The experimental analysis shows that the class adaptive structure element has a more prominent Abdou-pratt quality factor,peak signal-to-noise ratio,and structural similarity,and minor deviation of mean square and gradient magnitude similarity when morphological operations are performed on images,which indicates that it has higher accuracy of icon edge localization and retention of practical image details.In addition,the mean square and gradient magnitude similarity deviations are relatively small.The peak signal-to-noise ratio and structural similarity values are relatively large when using it for different images for erosion and expansion operations,indicating the effect of processing different images is close.However,the degree of weakening image details in light and dark is relatively tiny.（4）The research of watershed image segmentation method based on adaptive structural elementsIn order to solve the over-segmentation problem in traditional watershed image segmentation,the adaptive structure element is applied to watershed image segmentation.First,the adaptive structure element is constructed using the image’s local grayscale property and edge property,and the morphological gradient of the image is obtained using this structural element.Then,the gradient image is corrected by combining L₀ parametric gradient minimization and morphological open-closed hybrid reconstruction.Finally,the corrected gradient image is subjected to watershed segmentation to achieve effective extraction of image targets.The experimental analysis shows that the use of adaptive structural elements to correct the gradient image reduces the invalid local minima in the gradient image,effectively suppresses the generation of the over-segmentation phenomenon,improves the segmentation accuracy of the image target,and has a better segmentation effect.