Least Absolute Deviation Cut

The paper discussed a new additive extension of minimum cut by simultaneously minimizing intra cluster similarity bias and inter cluster similarity, Least Absolute Deviation Cut (LAD cut). The LAD cut can be proved convergent in finite iterative steps, and its theoretical conditions that the LAD cut can work well is also presented. Furthermore, its computational complexity is also analyzed. Numerical experimental results show that LAD cut may be useful for image segmentation.