Unsupervised active contour model for multiphase inhomogeneous image segmentation

This paper presents an unsupervised active contour model for multiphase inhomogeneous image segmentation. We propose the new model based on a local intensity fitting term and a phase balancing term by incorporating the idea of the local intensity fitting energy model into the phase balancing model. Instead of using intensity average constants, we use local fitting functions to approximate the intensities in different phases, thus the new model can segment inhomogeneous images. Besides, the new model can identify the number of phases automatically without any user input with the phase balancing term. Then a fast brute-force algorithm is provided to minimize the new nonlinear energy functional directly without computing the Euler-Lagrange equation. The new model has been applied to real images. Numerical results have demonstrated that the new model can deal with inhomogeneous images and give a reasonable number of phases simultaneously.