Image Segmentation by Unsupervised Graph Cuts
Image segmentation is a fundamental problem in vision and graphics. It is closely related to other vision and graphics researches such as object recognition or modeling. A number of studies have been conducted by many researches for a long time in the literature, but image segmentation is still remained as a challenging problem since it is inseparably connected to human perception.
In this research, we propose a novel approach to image segmentation based on combinatorial graph cut techniques, which finds an optimal segmentation of a given image, i.e., optimal segments and segment boundaries, by regarding it as a minimum cut problem in a weighted graph. Contrary to previous graph cut-based segmentation approaches, our method computes s-t minimum cuts using split moves in an unsupervised way. Thus, our method is called unsupervised graph cuts (see Figs. 1 and 2). The number of segments is determined during the split moves without user interaction.
Fig. 1 An illustration of unsupervised graph cuts for image segmentation. Solid red lines are s-t minimum cuts corresponding to segmentation boundaries.
Fig. 2 Overall flow of the image segmentation by unsupervised graph cuts. Color, texture, or motion images can be used as an image input of our method.
Results of color segmentation on the Berkeley dataset
Results of texture segmentation on the MIT VisTex dataset and the Berkeley dataset
Results of motion segmentation
 Jong-Sung Kim and Ki-Sang Hong, ˇ°Multiple Active Contours Using Divisive Graph Cuts,ˇ± In Proc. Korea Signal Processing Conference, 2006
 Jong-Sung Kim and Ki-Sang Hong, ˇ°Graph Cut-based Multiple Active Contours without Initial Contours and Seed Points,ˇ± In Proc. Image Processing and Image Understanding Workshop, 2007
 Jong-Sung Kim and Ki-Sang Hong, "A New Graph Cut-Based Multiple Active Contour Algorithm without Initial Contours and Seed Points," Machine Vision and Applications, 2007
 Jong-Sung Kim and Ki-Sang Hong, "Color-Texture Segmentation Using Unsupervised Graph Cuts," In Proc. 7th POSTECH-KYUTECH Joint Workshop on Neuroinformatics, 2007
 Jong-Sung Kim and Ki-Sang Hong, "Color-Texture Segmentation Using Unsupervised Graph Cuts," submitted to Pattern Recognition