xiehq / COSR

Content-oriented Sparse Representation (COSR) Denoising in CT Images with the Ability of Texture and Edge Preserving

Content‐oriented Sparse Representation (COSR) for CT Denoising with Preservation of Texture and Edge

This is a demostration MATLAB code package for Content-oriented Sparse Representation (COSR) Denoising in CT Images.

The COSR denoising method can effectively preserve the noise texture and image edges while reducing the strength of CT image noises.

It is free to use, distribute, modify and share the package for research and personal use; the only requirement is proper referencing to the authors.

Some result images

The COSR denoising method compared to the original SR method and others with a water phantom:

And compared with a pediatric head image (see our paper for details):

How to install and run the demo codes

(Tested on Windows x64 machines, please report any issue if it does not work in other arch/OS)

• Download or check out the COSR codes

• Navigate your MATLAB to the COSR folder

• run Step1_setup_COSR_denoising.m

  • Optional: if using Windows x64 and want to use pre-compiled binaries, comment out Line 7: compile_spams_cosrdenoising; in file compile_and_setup_spams.m.

• run Step2_sparsecoding_denoising_2D.m for 2D image denoising test

• run Step3_sparsecoding_denoising_3D.m for 3D image denoising test

If it does not work

• Make sure a compiler is set up in your MATLAB. run mex -setup. If a compiler is not set up, make sure you download one (some are free) and run mex -setup again.

• Read spams-matlab-v2.6\HOW_TO_INSTALL.txt and modify the spams-matlab-v2.6\compile_spams_COSR.m file according to your current OS.

Change parameter settings of the COSR denoising

• Some basic settings can be changed right inside the Step2_sparsecoding_denoising_2D.m and Step3_sparsecoding_denoising_3D.m files.

• More settings are in: COSR\sparsecoding_denoising_2D_paramSettings.m and COSR\sparsecoding_denoising_3D_paramSettings.m

• Please refer to our Medical Physics paper for details of these paramters

• Paramters related to the dictinary learning and OMP tools can be found inside: spams-matlab-v2.6\doc_spams_2.6.pdf

Contact

If you want to contact us for other reasons, please send us an email to

xiehuiqiao[@]gmail.com

Referencing COSR Denoising

If you use COSR in any publications, please reference the following papers:

Content‐oriented Sparse Representation (COSR) for CT Denoising with Preservation of Texture and Edge Huiqiao Xie, Tianye Niu, Shaojie Tang, Xiaofeng Yang, Nadja Kadom, Xiangyang Tang Medical Physics, 2018, Accepted Author Manuscript

Content-oriented sparse representation (COSR) denoising in CT images Huiqiao Xie, Nadja Kadom, Xiangyang Tang SPIE Medical Imaging 2018, Houston, Texas, United States, 10 - 15 February 2018 Presentation, Proceeding

References:

This COSR denoising demo code package uses mexCombinePatches, mexExtractPatches, mexOMP and mexTrainDL of the SPAMS package.

Just for avoiding any possible compatibility problems of further SPAMS releases, SPAMS v2.6 is enclosed with this COSR denoising demo.

• J. Mairal, F. Bach, J. Ponce and G. Sapiro. Online Learning for Matrix Factorization and Sparse Coding. Journal of Machine Learning Research, volume 11, pages 19-60. 2010.
• J. Mairal, F. Bach, J. Ponce and G. Sapiro. Online Dictionary Learning for Sparse Coding. International Conference on Machine Learning, Montreal, Canada, 2009
• http://spams-devel.gforge.inria.fr/