Color image segmentation algorithm based on local priority

A novel method that provides effective and robust segmentation of color images was designed. The image was pre-segmented by using the mean shift (MS) algorithm to form some segmented regions which could preserve the discontinuity characteristics of the image. Then, the similarity graph of the pre-segmented regions was constructed and the spectral clustering ensemble method was used to perform globally optimized clustering. To avoid some inappropriate partitioning when constructing similarity graph by only using the lightness information of pixels, a variable named local priority which considered both the lightness and detail information of the local region pixels was defined. The experimental results in color images verify the superiority of the designed method.