Spectral clustering has been widely used for image segmentation recently. There are certain issues when using spectral clustering for image segmentation, such as a high complexity. Moreover, it commonly similarity measure is a Gaussian kernel function. However, spectral clustering is very sensitive to the scale parameters in this similarity measure, which is difficult to determine a suitable parameter. For these problems, a modified superpixel segmentation method and a new similarity measure for improving Ng-Jordan-Weiss (NJW) method are presented in this study. Then the improved NJW method is applied to image segmentation. In the authors’ scheme, their modified superpixel segmentation method will be utilised to divide the image into several small regions, which are called superpixels. Then, the NJW method is used to cluster these superpixels into some meaningful regions. In NJW, the similarity between two adjacent superpixels is measured by a kernel fuzzy similarity measure. The improving NJW method for image segmentation not only has lower complexity but also not sensitivity to scale parameters. Experimental results have demonstrated that their method visible improvement both in diminishing segmentation error, and also it has a higher efficiency.
Read full abstract