Affiliations: ICT Research Center, Key Laboratory of Optoelectronic
Technology and System of the Education Ministry of China, Chongqing University,
Chongqing, China | College of Mechanical Engineering, University of South
China, Hengyang, Hunan, China | College of Mathematics and Statistics, Chongqing
University, Chongqing, China
Note: [] Corresponding author: Li Zeng, ICT Research Center, Key
Laboratory of Optoelectronic Technology and System of the Education Ministry of
China, Chongqing University, Chongqing 400044, China. Tel.: +86 23 8639 4828;
Fax: +86 23 6510 3562; E-mail: [email protected]
Abstract: Intensity inhomogeneity may cause considerable difficulties in
segmentation of CT image. In order to overcome the difficulties caused by
intensity inhomogeneity, the region-scalable fitting (RSF) model was put
forward. RSF model draws upon intensity information in local regions with a
controllable scale. But only using intensity information may lead to slow
convergence rate and poor denoise ability. Combining the method of robust
statistics, RSF model is improved in this paper. In the improved model, the
intensity in RSF model is replaced with local robust statistics which is the
weighted combination of inter-quartile range, mean absolute deviation and
intensity median in local region. Inter-quartile range and mean absolute
deviation in local region are introduced to sharpen object boundaries, and
intensity median in local region is introduced to reduce image noise. The
contrast experiments between RSF model and the improved model are provided,
which demonstrate the fast convergence rate and robustness to noise of the improved model.
