Eidors-logo    

EIDORS: Electrical Impedance Tomography and Diffuse Optical Tomography Reconstruction Software

EIDORS (mirror)
Main
Documentation
Tutorials
− Image Reconst
− Data Structures
− Applications
− FEM Modelling
− GREIT
− Old tutorials
Workshop
Download
Contrib Data
GREIT
Browse Docs
Browse SVN

News
Mailing list
(archive)
FAQ
Developer
                       

 

Hosted by
SourceForge.net Logo

 

Compare selection of hyperparameters for Total Variation

Total Variation typically uses a GN step to initialize the first iteration. We thus have a hyperparameter for the GN step (α1), and another for the GN interations (α2).

Simluation object

Simulate shape with edges
% TV: Create object $Id: TV_hyperparams01.m 3365 2012-07-02 08:05:40Z aadler $

imb=  mk_common_model('c2c2',16);
img= mk_image(imb.fwd_model, 1);

vh= fwd_solve( img );

img.elem_data([25,37,49:50,65:66,81:83,101:103,121:124])=2;
img.elem_data([95,98:100,79,80,76,63,64,60,48,45,36,33,22])=2;
    
vi= fwd_solve( img );

subplot(221)
show_fem(img);
axis equal; axis off;
print_convert TV_hyperparams01a.png '-density 100';


Figure: Shape that is to be reconstructed (generated on 576 element mesh)

Create TV reconstruction model

% TV: Reconstruction model $Id: TV_hyperparams02.m 3428 2012-07-02 20:56:41Z bgrychtol $

maxit=20;  % max number of iterations
imdl=mk_common_model('b2c0',16);

invtv= eidors_obj('inv_model', 'EIT inverse');
invtv.reconst_type= 'difference';
invtv.jacobian_bkgnd.value= 1;

invtv.fwd_model=                  imdl.fwd_model;
invtv.solve=                      @inv_solve_TV_pdipm;
invtv.R_prior=                    @prior_TV;
invtv.parameters.term_tolerance=  1e-6;
invtv.parameters.keep_iterations= 1;
invtv.parameters.max_iterations=  maxit;

subplot(221)
show_fem(invtv.fwd_model);
axis equal; axis off;
print_convert TV_hyperparams02a.png '-density 100';




Figure: Reconstruction model (256 elements)

Reconstruction with no noise

% Iterate over params $Id: TV_hyperparams03.m 3428 2012-07-02 20:56:41Z bgrychtol $
name_base = 'tv_hp_00_img-a1=%3.1f-a2=%3.1f.jpg';
alpha1list= [1.5:0.5:4.5];
alpha2list= [3.5:1.0:9.5];
for alpha2= alpha2list
   for alpha1= alpha1list
      invtv.hyperparameter.value =   10^-alpha2;
      invtv.inv_solve_TV_pdipm.alpha1= 10^-alpha1;
      imgs= inv_solve(invtv,vh,vi);
      imgs.elem_data= imgs.elem_data(:,[1,2,5,maxit]);
      imgs.calc_colours.window_range=.2;
      imgs.calc_colours.clim=1.2;
      out_img= show_slices(imgs);
      imwrite(out_img, colormap, sprintf(name_base,alpha1,alpha2) );
    end
end

To display these results, create an html table with matlab

% Generate HTML frame to view $Id: TV_hyperparams04.m 2639 2011-07-12 07:39:06Z bgrychtol $

fid= fopen('TV-params-NSR=0.html','w');

a=sprintf('%calpha;',38); % alpha
m=sprintf('%cminus;',38); % alpha
s=sprintf('%c',60); % less than 
e=sprintf('%c',62); % greater than
tr= [s,'TR',e]; etr= [s,'/TR',e];
th= [s,'TH',e]; eth= [s,'/TH',e];
td= [s,'TD',e]; etd= [s,'/TD',e];
sub=[s,'SUB',e];esub=[s,'/SUB',e];
sup=[s,'SUP',e];esup=[s,'/SUP',e];


fprintf(fid,[s,'TABLE',e,tr,th]);
for alpha1= alpha1list
   fprintf(fid,[th,a,sub,'1',esub,'=10',sup,m,'%3.1f',esup],alpha1);
end
fprintf(fid,'\n');

for alpha2= alpha2list
   fprintf(fid,['\n',tr,'\n',th,a,sub,'2',esub,'=10', ...
                sup,m,'%3.1f',esup,'\n'],alpha2);
   for alpha1= alpha1list
      fprintf(fid,[td,s,'img src="%s"',e,'\n'], sprintf(name_base,alpha1,alpha2) );
   end
end

fprintf(fid,['\n',s,'/TABLE',e,'\n']);
fclose(fid);

Reconstruction Results (No Noise) (GN param = α1, TV param= α2)

α1=10−1.5α1=10−2.0α1=10−2.5α1=10−3.0α1=10−3.5α1=10−4.0α1=10−4.5
α2=10−3.5
α2=10−4.5
α2=10−5.5
α2=10−6.5
α2=10−7.5
α2=10−8.5
α2=10−9.5

Reconstruction with 20db SNR noise

Add Noise

% Add noise $Id: TV_hyperparams05.m 1535 2008-07-26 15:36:27Z aadler $

sig= sqrt(norm(vi.meas - vh.meas));
m= size(vi.meas,1);
vi.meas = vi.meas + .01*sig*randn(m,1);

Calculations

% Iterate over params $Id: TV_hyperparams06.m 3428 2012-07-02 20:56:41Z bgrychtol $
name_base= 'tv_hp_20_img-a1=%3.1f-a2=%3.1f.jpg';

alpha1list= [1.5:0.5:4.5];
alpha2list= [3.5:1.0:9.5];
for alpha2= alpha2list
   for alpha1= alpha1list
      invtv.hyperparameter.value =   10^-alpha2;
      invtv.inv_solve_TV_pdipm.alpha1= 10^-alpha1;
      imgs= inv_solve(invtv,vh,vi);
      imgs.elem_data= imgs.elem_data(:,[1,2,5,maxit]);
      imgs.calc_colours.window_range=.2;
      imgs.calc_colours.clim=1.2;
      out_img= show_slices(imgs);
      imwrite(out_img, colormap, sprintf(name_base,alpha1,alpha2) );
    end
end

To display these results, create an html table with matlab

% Generate HTML frame to view $Id: TV_hyperparams07.m 2733 2011-07-13 21:04:58Z anon-eidors $

fid= fopen('TV-params-NSR=0.01.html','w');


a=sprintf('%calpha;',38); % alpha
m=sprintf('%cminus;',38); % alpha
s=sprintf('%c',60); % less than 
e=sprintf('%c',62); % greater than
tr= [s,'TR',e]; etr= [s,'/TR',e];
th= [s,'TH',e]; eth= [s,'/TH',e];
td= [s,'TD',e]; etd= [s,'/TD',e];
sub=[s,'SUB',e];esub=[s,'/SUB',e];
sup=[s,'SUP',e];esup=[s,'/SUP',e];


fprintf(fid,[s,'TABLE',e,tr,th]);
for alpha1= alpha1list
   fprintf(fid,[th,a,sub,'1',esub,'=10',sup,m,'%3.1f',esup],alpha1);
end
fprintf(fid,'\n');

for alpha2= alpha2list
   fprintf(fid,['\n',tr,'\n',th,a,sub,'2',esub,'=10', ...
                sup,m,'%3.1f',esup,'\n'],alpha2);
   for alpha1= alpha1list
      fprintf(fid,[td,s,'img src="%s"',e,'\n'], sprintf(name_base,alpha1,alpha2) );
   end
end

fprintf(fid,['\n',s,'/TABLE',e,'\n']);
fclose(fid);

Reconstruction Results (20dB SNR) (GN param = α1, TV param= α2)

α1=10−1.5α1=10−2.0α1=10−2.5α1=10−3.0α1=10−3.5α1=10−4.0α1=10−4.5
α2=10−3.5
α2=10−4.5
α2=10−5.5
α2=10−6.5
α2=10−7.5
α2=10−8.5
α2=10−9.5

Last Modified: $Date: 2017-02-28 13:12:08 -0500 (Tue, 28 Feb 2017) $ by $Author: aadler $