Image segmentation method and system based on cluster-shaped boundary closure clustering

Abstract

The invention discloses an image segmentation method and system based on cluster-shaped boundary closure clustering. The method comprises the steps of S1, inputting an original data set of a to-be-segmented source image and eliminating edge noisy points; s2, extracting boundary points of a cluster from the data set after the edge noise points are eliminated; s3, forming a boundary closure of the cluster by using a boundary point surrounding method according to the extracted boundary points of the cluster; and S4, determining a clustering number according to the obtained number of the boundary closures of the cluster, performing extension clustering on non-boundary points in the original data set from the boundary to the center, and realizing segmentation of the to-be-segmented source image according to a clustering result. The method has the advantages of simple implementation method, high segmentation efficiency and precision, good segmentation effect and the like.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image segmentation method and system based on cluster shape boundary closure clustering. Background technique [0002] Image segmentation is to use the gray scale, color, texture, shape and other characteristics of the image to divide the image into several non-overlapping areas, and make these features similar in the same area, and there are obvious differences between different areas. sex. Image segmentation technology has been widely used in various fields. For example, in the field of locomotive inspection, it can be applied to the segmentation of hub crack images to detect cracks in time and ensure driving safety; in biomedical engineering, the segmentation of liver CT images can be used for clinical Aids in therapy and pathology research. [0003] The key technology in image segmentation is clustering. For image segmentation, two types of clustering methods b...

AI Technical Summary

Problems solved by technology

Finding cluster centers is more convenient in convex data sets, but it is often difficult to find cluster centers in non-convex data sets, because the method of minimizing the objective function is usually only applicable to convex data sets

Therefore, such as the K-means algorithm, there will be problems such as being sensitive to the number of clusters K, sensitive to outliers, and outliers have a great influence on the calculation of the centroid, and sensitive to the initial point randomly selected at the beginning of the algorithm.

[0005] (2) Density-based clustering methods, such as the DBSCAN algorithm, need to artificially set the neighborhood radius Eps and the number of samples MinPts in the neighborhood, so after setting these two parameters, the density of the sample cluster will be limited accordingly, so the cluster Although the class method can find clusters with complex shapes, it is more sensitive to the density of clusters. For example, the DBSCAN algorithm is particularly sensitive to the parameters MinPts and Eps, so the clustering effect on clusters with uneven density distribution is poor.

[0006] To sum up, the traditional partition clustering method and the density-based clustering method in image segmentation usually start from the local characteristics of the data distribution and extend from the center, which will ignore the overall characteristics of the data distribution and affect the clustering. Class effect; and traditional clustering methods usually need to manually specify the number of clusters, but in fact it is difficult to know the exact number of clusters in the sample space in advance, which makes the clustering process of image segmentation very dependent on the prior parameters setting, resulting in poor image segmentation

Images