Fast and Robust Segmentation of Low Contrast Biomedical Images
Jan Hubený, Pavel Matula · 2006
This paper presents fast and robust algorithm for minimization of Chan--Vese energy functional. Proposed technique is based on recently published k-Means level set and threshold dynamics approximations of Chan--Vese functional. The approximation algorithms are combined in order to preserve their individual advantages and avoid their limitations. Hence, the proposed hybrid algorithm is robust and converges reasonably fast to steady state and is suitable for two-phase segmentation of low contrast biomedical data. A simple numerical scheme for threshold dynamics method is derived in the paper. Results of the hybrid algorithm that are better than results of both k-Means level set and threshold dynamics methods employed individually are presented.