%FILTIM Fixed mapping to filter images in datasets and datafiles
%
% B = FILTIM(A,COMMAND,{PAR1,PAR2,....},SIZE)
% B = A*FILTIM([],COMMAND,{PAR1,PAR2,....},SIZE)
% B = A*FILTIM(COMMAND,{PAR1,PAR2,....},SIZE)
%
% INPUT
% A Dataset or datafile with multi-band image objects
% COMMAND String with function name of command to be executed
% {PAR1,...} Cell array with optional parameters to COMMAND
% SIZE Output size of the mapping (default: input size)
%
% OUTPUT
% B Dataset containing multi-band images processed by
% COMMAND, band by band.
%
% DESCRIPTION
% For each band of each image in A a filter operation is performed
%
% OBJECT_OUT = COMMAND(OBJECT_IN,PAR1,PAR2,....)
%
% Object images as well as feature images are supported.
% The results are collected and stored in B. In case A (and thereby B) is
% a datafile, execution is postponed until conversion into a dataset, or a
% call to SAVEDATAFILE.
%
% The difference between FILTIM and the similar command FILTM is that
% FILTIM is aware of the band structure of the objects. As FILTIM treats
% the bands separately it cannot be used for commands that change the number
% of bands (like RGB2GRAY) or need to access them all simultaneaously.
% However, FILTM can handle cells and FILTIM cannot.
%
% EXAMPLE
% A = delft_images; B = A([120 121 131 230])*doublem*col2gray;
% E = B*filtim('fft2')*filtim('abs')*filtim('fftshift');
% figure; show(E); figure; show((1+E)*filtim('log'));
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, DATAFILES, IM2OBJ, IM2FEAT, FILTM, DATA2IM, PROCM, MAPM
% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
function b = filtim(varargin)
if nargin == 0
error('No command found')
end
mapname = 'ImageFilt';
% check for mapping definition: first argument empty or string
varargin = shiftargin(varargin,'char');
if isempty(varargin{1})
b = prmapping(mfilename,'fixed',varargin(2:end));
b = setname(b,mapname);
% set size_out if user wants so
if numel(varargin) == 4, b = setsize_out(b,varargin{4}); end
return
end
% now we have: proc(data,command,pars)
% data might be dataset, datafile, cell array, double, structure array
nout = max(1,nargout);
argin = setdefaults(varargin,[],[],[],[]);
[a,command,pars,outsize] = deal(argin{:}); % use nice names
if isempty(pars)
pars = {};
elseif ~iscell(pars)
pars = {pars};
end
if isdatafile(a) % for datafiles filters are stored
isobjim(a);
if isempty(getpostproc(a))&& ~ismapping(command)
% as preprocessing (if no postproc defined)
b = addpreproc(a,mfilename,{command pars},outsize);
else % or as mapping as postprocessing
if ismapping(command) % we have already a mapping
v = command;
else % just a string, construct mapping
v = prmapping(mfilename,'fixed',{command,pars});
v = setname(v,mapname);
end
if ~isempty(outsize)% user wants to set an output size (hope he has good reasons)
v = setsize_out(v,outsize); % add it to mapping
end
b = addpostproc(a,v); % store mapping
end
elseif isdataset(a)
% convert to image and process
if isobjim(a)
m = size(a,1);
d = data2im(a);
out = feval(mfilename,d,command,pars); % effectively jumps to double (below)
out = double(out);
fsize = size(out);
if m > 1
fsize = fsize(1:3);
if fsize(3) == 1
fsize = fsize(1:2);
end
end
% store processed images in dataset
b = setdata(a,im2obj(out,fsize));
b = setfeatsize(b,fsize);
elseif isfeatim(a)
k = size(a,2);
d = data2im(a);
out = feval(mfilename,d,command,pars); % effectively jumps to double (below)
out = double(out);
% store processed images in dataset, assume size didn't change
b = setdata(a,im2feat(out,getobjsize(a)));
% b = setobjsize(b,osize);
else
error('No images found in dataset')
end
else % double
% make imsize 4D: horz*vert*bands*objects
imsize = size(a);
if length(imsize) == 2
imsize = [imsize 1 1];
elseif length(imsize) == 3
imsize = [imsize 1];
end
% find size of first image
first = execute(command,getsubim(a,1,1),pars);
%first = feval(command,getsubim(a,1,1),pars{:});
if all(imsize(3:4) == 1) % preserve DipLib format for the time being
b = first;
return
end
first = double(first); % for Dip_Image users
outsize = size(first);
if length(outsize) == 3 % single subimage generates multiple bands !!
b = repmat(first(:,:,1),[1 1 imsize(3)*outsize(3) imsize(4)]);
else
b = repmat(first,[1 1 imsize(3:4)]);
end
% process all other images
[nn,s,count] = prwaitbarinit('Filtering %i images',imsize(4));
for j = 1:imsize(4)
for i = 1:imsize(3)
ima = double(execute(command,getsubim(a,i,j),pars));
%ima = double(feval(command,getsubim(a,i,j),pars{:}));
if (any(outsize ~= size(ima))) % check size
error('All image sizes should be the same')
end
if length(outsize) == 2 % simple case: image_in --> image_out
b(:,:,i,j) = ima;
else % advanced: image_in --> bands out
b(:,:,i:imsize(3):end,j) = ima;
end
end
count = prwaitbarnext(imsize(4),'Filtering %i images: ',count);
end
end
return
function asub = getsubim(a,i,j)
% needed as Dip_Image cannot handle 4d subsript if data is 2d or 3d
n = ndims(a);
if n == 2
if i ~= 1 | j ~= 1
error('Illegal image requested')
end
asub = a;
elseif n == 3
if j ~= 1
error('Illegal image requested')
end
asub = a(:,:,i);
elseif n == 4
asub = a(:,:,i,j);
else
error('Incorrect image supplied')
end
function out = execute(command,a,pars)
exist_command = exist(command);
if isstr(command) & any([2 3 5 6] == exist_command)
if isempty(pars) | numel(pars) == 1 & isempty(pars{:})
out = feval(command,a);
else
out = feval(command,a,pars{:});
end
elseif ismapping(command)
out = prmap(a,command);
else
error([command ': Filter command not found'])
end
return