IJMLC 2015 Vol. 5(3): 198-205 ISSN: 2010-3700
DOI: 10.7763/IJMLC.2015.V5.507

Mutiscale Texture Segmentation Using Contextual Hidden Markov Tree Models

Din-Chang Tseng and Ruei-Lung Chen

Abstract—A multiscale texture segmentation approach based on contextual hidden Markov tree (CHMT) model and boundary refinement is proposed. A hidden Markov tree (HMT) model is a probabilistic model for capturing persistence properties of wavelet coefficients without considering clustering properties. We have proposed the CHMT model to enhance the clustering properties by adding extended coefficients associated with wavelet coefficients in every scale. In this study, we train the CHMT parameters for every texture and then use them to compute maximum likelihoods for every dyadic square region at every scale in an image which will be segmented. Then the boundary refinement algorithm is adopted to fuse the different-scale segmented results to improve the final results. We demonstrate the performance of the proposed method on synthetic and aerial images; moreover, the comparison with other methods is also provided to show the effectiveness of the proposed method.

Index Terms—Contextual hidden Markov tree model, maximum likelihood, multiscale texture image segmentation, wavelet transform.

Din-Chang Tseng and Ruei-Lung Chen are with the Institute of Computer Science and Information Engineering, National Central University, Jhongli, 32001 Taiwan (e-mail: tsengdc@ip.csie.ncu.edu.tw, rlchen17@gmail.com).

[PDF]

Cite: Din-Chang Tseng and Ruei-Lung Chen, "Mutiscale Texture Segmentation Using Contextual Hidden Markov Tree Models," International Journal of Machine Learning and Computing vol. 5, no. 3, pp. 198-205, 2015.

General Information

  • ISSN: 2010-3700 (Online)
  • Abbreviated Title: Int. J. Mach. Learn. Comput.
  • Frequency: Bimonthly
  • DOI: 10.18178/IJMLC
  • Editor-in-Chief: Dr. Lin Huang
  • Executive Editor:  Ms. Cherry L. Chen
  • Abstracing/Indexing: Scopus (since 2017), EI (INSPEC, IET), Google Scholar, Crossref, ProQuest, Electronic Journals Library.
  • E-mail: ijmlc@ejournal.net