Sonar Image Segmentation Based On Rough Set

Image segmentation is a technique for extracting the regions of internet from the image,which is essential for correct recognition of final target.Image segmentation,as a difficult and hot spot for sonar image processing,has received great attention from researchers.In this thesis,rough set theory is used to segment sonar images.Rough set theory is an effective method for solving incomplete,uncertain and inaccurate problems.For dealing with sonar image segmentation problems lacking prior knowledge,rough set theory provides a very good solution.The main contents of this thesis are as follows:(1)In image segmentation,this thesis proposes to use the standard deviation,Euclidean distance and maximum gradient as the main features to determine the edge of the sonar image.The standard deviation,the Euclidean distance and the maximum gradient are used to construct the indiscernibility relation of the rough set of the single feature,the double feature and the triple feature respectively,and the influence of different characters constructions' indiscernibility relation on segmentation results are discussed by experiments.Indiscernibility relation is used to divide the equivalence classes.(2)Based on hypothesis testing,an adaptive region fusion method is proposed.The mean value of variance in the horizontal and vertical directions of the image is used as an important criterion for the fusion of equivalence classes.The number of pixels in the equivalence class of adjacent labels is the condition of t-test or Z-test equivalence class fusion.After fusion,the equivalence class is given.The new label is re-assigned.When the label is no longer changed,it is regarded as the end of the fusion.The isolated region of the image is removed,and the equivalence class is assigned a different color.As a result,the final segmented image is obtained.The algorithm of this thesis is mainly used for sonar images,and has achieved significant effects of image segmentation.