/normxcorr/trunk

function [validx,validy]=automate_image(grid_x,grid_y,filenamelist,validx,validy);

% Code to start actual image correlation

% Programmed by Chris and Rob

% Last revision: 09/10/08

% The automation function is the central function and processes all markers and 

% images by the use of the matlab function cpcorr.m. 

% Therefore the Current directory in matlab has to be the folder where 

%  automate_image.m finds the filenamelist.mat, grid_x.dat and grid_y.dat as well 

% as the images specified in filenamelist.mat. Just type automate_image; and 

% press ENTER at the command line of matlab. 

% At first, automate_image.m will open the first image in the filenamelist.mat and 

% plot the grid as green crosses on top. The next step will need some time since 

% all markers in that image have to be processed for the first image. After correlating 

% image one and two the new raster positions will be plotted as red crosses. On top 

% of the image and the green crosses. The next dialog will ask you if you want to 

% continue with this correlation or cancel. If you press continue, automate_image.m 

% will process all images in the filenamelist.mat. The time it will take to process 

% all images will be plotted on the figure but can easily be estimated by knowing the 

% raster point processing speed (see processing speed). 

% Depending on the number of images and markers you are tracking, this process 

% can take between seconds and days. For 100 images and 200 markers a decent 

% computer should need 200 seconds. To get a better resolution you can always 

% run jobs overnight (e.g. 6000 markers in 1000 images) with higher resolutions. 

% Keep in mind that CORRSIZE which you changed in cpcorr.m will limit your 

% resolution. If you chose to use the 15 pixel as suggested a marker distance of 

% 30 pixel will lead to a full cover of the strain field. Choosing smaller marker 

% distances will lead to an interpolation since two neighboring markers share 

% pixels. Nevertheless a higher marker density can reduce the noise of the strain field.

% When all images are processed, automate_image will write the files validx.mat, 

% validy.mat, validx.txt and validy.txt. The text files are meant to store the result in a 

% format which can be accessed by other programs also in the future.

% Changed by Suren A. Chilingaryan <csa@dside.dyndns.org> to optimize performance

% by tighter integration with Matlab images toolkit (few sources from the toolkit

% are moved to the current source tree and adjusted to benefit from knowledge of 

% tasks we are solving here. As well CUDA framework, if available, is used to

% speed-up core computations using parallel processors of latest NVidia video

% cards.

% OPTIMIZE parameter is controlling which optimizations should be used. 

%  0 - The original version, no optimizations

%  1 - The modified sources from images toolkit are used

%  2 - The CUDA matlab plugin is used to compute FFT's

%  3 - Most of computations are performed on NVidia card

% You can safely set an optimization level to 3. If NVidia card is not available

% the level will be automatically reduced to 1.

OPTIMIZE = 3;

CORRSIZE = 15;	

PRECISION = 1000;

SILENT = 1;

% exist('grid_x')

% exist('grid_y')

% exist('filenamelist')

% exist('validx')

% exist('validy')

if exist('data', 'dir')

    datadir = 'data/';

else

    datadir = '';

end

if exist('images', 'dir')

    imagedir = 'images/';

else

    imagedir = '';

end

% Load necessary files

if exist('grid_x')==0

    load([datadir, 'grid_x.dat'])              % file with x position, created by grid_generator.m

end

if exist('grid_y')==0

    load([datadir, 'grid_y.dat'])              % file with y position, created by grid_generator.m

end

if exist('filenamelist')==0

    load([datadir, 'filenamelist'])            % file with the list of filenames to be processed

end

resume=0;

if exist('validx')==1

    if exist('validy')==1

        resume=1;

        [Rasternum Imagenum]=size(validx);

    end

end

% Initialize variables

input_points_x=grid_x;

base_points_x=grid_x;

input_points_y=grid_y;

base_points_y=grid_y;

if resume==1

    input_points_x=validx(:,Imagenum);

    input_points_y=validy(:,Imagenum);

    inputpoints=1;

end

[row,col]=size(base_points_x);      % this will determine the number of rasterpoints we have to run through

[r,c]=size(filenamelist);           % this will determine the number of images we have to loop through

% Open new figure so previous ones (if open) are not overwritten

if (~SILENT)

    h=figure;

    imshow([imagedir, filenamelist(1,:)])           % show the first image

    title('Initial Grid For Image Correlation (Note green crosses)')        % put a title

    hold on

    plot(grid_x,grid_y,'g+')            % plot the grid onto the image

    hold off

    % Start image correlation using cpcorr.m

    g = waitbar(0,sprintf('Processing images'));        % initialize the waitbar

    set(g,'Position',[275,50,275,50])                   % set the position of the waitbar [left bottom width height]

end

firstimage=1;

if resume==1

    firstimage=Imagenum+1

end

%Initializing hardware code

base = uint8(mean(double(imread([imagedir, filenamelist(1,:)])),3));            % read in the base image ( which is always  image number one. You might want to change that to improve correlation results in case the light conditions are changing during the experiment

base_points_for(:,1)=reshape(base_points_x,[],1);

base_points_for(:,2)=reshape(base_points_y,[],1);

%some checks (moved here from cpcorr)

if any(base_points_for(:)<0.5) || any(base_points_for(:,1)>size(base,2)+0.5) || any(base_points_for(:,2)>size(base,1)+0.5)

    msg = sprintf('In function %s, Control Points must be in pixel coordinates.',mfilename);

    eid = sprintf('Images:%s:cpPointsMustBeInPixCoord',mfilename);

    error(eid, msg);

end

%moved from normxcorr2

% We can precalculate it here, since near-edge images are dropped and, therefore, A,T always have the same size 

% n is width and height of T, but following computations are done for A

% outsize is sum if widths and heights of T and A - 1 (outsize = size(A) + size(T) - 1)

outsize = 6 * CORRSIZE + 1;

n = 2*CORRSIZE + 1;

mn = n*n;

rects_base = dic_calc_rects(base_points_for,2*CORRSIZE,base);

data_base = struct;

ncp = size(base_points_for, 1);

if OPTIMIZE > 2

    hwid = normxcorr_hw();

    if hwid > 0

	err = normxcorr_hw(hwid, 1, ncp, CORRSIZE);

	if (err ~= 0)

	    normxcorr_hw(1);

	    OPTIMIZE = 2;

	end

    else

	if hwid < 0

	    OPTIMIZE = 1;

	else

	    OPTIMIZE = 2;

	end

    end

else

    hwid = 0;

end

if OPTIMIZE > 0

%  if OPTIMIZE > 2

%    crop_dim = [min(base_points_x(:)) - 2*CORRSIZE, min(base_points_y(:)) - 2 * CORRSIZE, max(base_points_x(:)) + 2*CORRSIZE, max(base_points_y(:)) + 2 * CORRSIZE]

%    crop_size = (crop_dim(3) - crop_dim(1)) * (crop_dim(4) - crop_dim(2))

%    mesh_size = ncp * 4 * CORRSIZE * CORRSIZE

%  end

  for icp = 1:ncp

    if isequal(rects_base(icp,3:4),[0 0])

	%near edge

	data_base(icp).skip = 1;

	continue

    end

    sub_base = imcrop(base,rects_base(icp,:));

    if any(~isfinite(sub_base(:)))

	%NaN or Inf

	data_base(icp).skip = 1;

	continue

    end

    data_base(icp).skip = 0;

    if (OPTIMIZE > 2)

	if sum(sub_base(:)) == sub_base(1)*numel(sub_base)

	    eid = sprintf('Images:%s:sameElementsInTemplate',mfilename);

	    msg = 'The values of TEMPLATE cannot all be the same.';

	    error(eid,'%s',msg);

	end

	normxcorr_hw(hwid, 10, icp, sub_base);

%{

	double_sub_base = double(sub_base);

	local_sum_A = dic_local_sum(double_sub_base,n,n);

	local_sum_A2 = dic_local_sum(double_sub_base.*double_sub_base,n,n);

	denom_A = sqrt( max(( local_sum_A2 - (local_sum_A.^2)/mn ) / (mn-1), 0) );

	i_nonzero = find(denom_A~=0);

	normxcorr_hw(hwid, 10, icp, sub_base, single(local_sum_A), single(denom_A), uint16(i_nonzero));

%}

    else

	double_sub_base = double(sub_base);

	if (numel(sub_base) < 2) ||  (std(double_sub_base(:)) == 0)

	    % This check is moved here for normxcorr2

	    eid = sprintf('Images:%s:sameElementsInTemplate',mfilename);

	    msg = 'The values of TEMPLATE cannot all be the same.';

	    error(eid,'%s',msg);

	end

	local_sum_A = dic_local_sum(double_sub_base,n,n);

	local_sum_A2 = dic_local_sum(double_sub_base.*double_sub_base,n,n);

	% Note: diff_local_sums should be nonnegative, but may have negative

	% values due to round off errors. Below, we use max to ensure the

	% radicand is nonnegative.

	denom_A = sqrt( max(( local_sum_A2 - (local_sum_A.^2)/mn ) / (mn-1), 0) );

	i_nonzero = find(denom_A~=0);

	data_base(icp).denom_A = denom_A;

	data_base(icp).i_nonzero = i_nonzero;

	data_base(icp).local_sum_A = local_sum_A;

	data_base(icp).sub_base = sub_base;

	if (OPTIMIZE > 1)

	    data_base(icp).sub_base_fft = fft2_cuda(double_sub_base, outsize, outsize);

	else

	    data_base(icp).sub_base_fft = fft2(double_sub_base, outsize, outsize);

	end

    end

    data_base(icp).base_fractional_offset = base_points_for(icp,:) - round(base_points_for(icp,:)*PRECISION)/PRECISION;

  end

  if (OPTIMIZE > 2)

     normxcorr_hw(hwid, 2);

  end

end

%normxcorr_hw(hwid);

%return;

for i=firstimage:(r-1)               % run through all images

    tic             % start the timer

    input = uint8(mean(double(imread([imagedir, filenamelist((i+1),:)])),3));       % read in the image which has to be correlated

    input_points_for(:,1)=reshape(input_points_x,[],1);         % we reshape the input points to one row of values since this is the shape cpcorr will accept

    input_points_for(:,2)=reshape(input_points_y,[],1);

    if OPTIMIZE > 0

	input_correl(:,:)=dic_cpcorr(CORRSIZE, PRECISION, OPTIMIZE, hwid, data_base, input_points_for, input);           % here we go and give all the markers and images to process to cpcorr.m which ic a function provided by the matlab image processing toolbox

    else

	input_correl(:,:)=cpcorr(input_points_for, base_points_for, input, base);

    end

    input_correl_x=input_correl(:,1);                                       % the results we get from cpcorr for the x-direction

    input_correl_y=input_correl(:,2);                                       % the results we get from cpcorr for the y-direction

    validx(:,i)=input_correl_x;                                                     % lets save the data

    savelinex=input_correl_x';

    dlmwrite([datadir, 'resultsimcorrx.txt'], savelinex , 'delimiter', '\t', '-append');       % Here we save the result from each image; if you are desperately want to run this function with e.g. matlab 6.5 then you should comment this line out. If you do that the data will be saved at the end of the correlation step - good luck ;-)

    validy(:,i)=input_correl_y;

    saveliney=input_correl_y';

    dlmwrite([datadir, 'resultsimcorry.txt'], saveliney , 'delimiter', '\t', '-append');

    % Update base and input points for cpcorr.m

    input_points_x=input_correl_x;

    input_points_y=input_correl_y;

    if (~SILENT) 

	waitbar(i/(r-1))                                % update the waitbar

	imshow([imagedir, filenamelist(i+1,:)])                     % update image

	hold on

	plot(grid_x,grid_y,'g+')                                % plot start position of raster

	plot(input_correl_x,input_correl_y,'r+')        % plot actual postition of raster

	hold off

	drawnow

	time(i)=toc;                                                 % take time

	estimatedtime=sum(time)/i*(r-1);            % estimate time to process

	title(['# Im.: ', num2str((r-1)),'; Proc. Im. #: ', num2str((i)),'; # Rasterp.:',num2str(row*col), '; Est. Time [s] ', num2str(round(estimatedtime)), ';  Elapsed Time [s] ', num2str(round(sum(time)))]);    % plot a title onto the image

	drawnow

    end

end    

if OPTIMIZE > 2

    normxcorr_hw(hwid);

end

if (~SILENT)

    close(g)

end

close all

% save

if (~SILENT)

    save ([datadir,'time.dat'], 'time', '-ascii', '-tabs');

end

save ([datadir,'validx.dat'], 'validx', '-ascii', '-tabs');

save ([datadir,'validy.dat'], 'validy', '-ascii', '-tabs');