Attribute Neighborhood Similarity And Soft Rough C-Means Image Edge Detection

Image edges contain rich information.Image edge detection can extract image structure,texture features,target contour and other information,and remove redundant information,which is the basis and premise of image analysis such as image segmentation,target recognition and region extraction.The existing edge detection algorithms are limited in practical application.It is the focus of current image processing research that how to achieve accurate,universal and robust image edge detection.To depict the edges of the image more accurately,Soft Rough C-Means edge detection based on Neighborhood Similarity of Spectral attributes(SRCM?NSSA)is proposed in this paper.The specific content is as follows,(1)To describe the similarity between the center pixel and the neighborhood pixels more accurately,the definition method of attribute neighborhood similarity was studied,the spectral attribute neighborhood similarity,membership neighborhood similarity and mixed neighborhood similarity were defined respectively.The spectral attribute neighborhood similarity is the mean of spectral attribute similarity in the pixel neighborhood,which can describe the correlation between the neighborhood pixel and the center pixel from the perspective of spectral attribute.Membership neighborhood similarity is the mean of membership similarity of the neighborhood pixels,which can describe the influence of neighborhood pixels on the center pixel from the perspective of membership.Mixed neighborhood similarity normalized the two attributes neighborhood similarity and combined it by linear weighting,so that the effect intensity of the two kinds of similarity could be changed by adjusting the weight.(2)Soft Rough C-Means edge detection based on Neighborhood Euclidean Distance(SRCM?NED)was designed by combining the Soft Rough set model and Rough C-Means clustering algorithm.The Euclidean distance between neighborhood pixels and clustering center is calculated and normalized.The soft upper approximations and soft lower approximations redefined by neighborhood Euclidean distance,and the clustering center calculation formula is updated by combining the positive domain and boundary domain divided by soft upper and lower approximations.The roughness calculation was redefined,and the number of iterations of soft roughness C-means algorithm was controlled by setting roughness threshold,so that the segmentation of boundary pixels by the algorithm was adjusted iteratively.(3)To improve the robustness of edge detection algorithm of SRCM,SRCM?NSSA was constructed by combining spectral attribute neighborhood similarity.Firstly,the neighborhood similarity of spectral attribute from pixel to cluster center is calculated.The soft upper and lower approximations were redefined by using the neighborhood similarity of spectral attributes,and new boundary domains and positive domains were obtained.Finally,the SRCM?NSSA algorithm was realized according to the process of SRCM edge detection algorithm.The paper include 41 figures,6 tables and 78 references.