peak_labelling.m

% written by Chris

function [validx,validy]=peak_labelling;

% The peak labelling function is an alternative methode to the
% automate_image function. The difference between the two functions is,
% that the automate_image function the correlation coefficient uses to
% track a fixed grid of markers, while the peak_labelling function is
% searching for peaks in a base image and tryes to fit a gauss function to
% it. Therefore the image should have a very low background and bright
% round peaks. If this is the case peak_labelling will find these peaks fit
% the gauss function in x and y direction to each of the peaks and then
% tracks all peaks in all images. The output files are fitxy.dat,
% validx.dat and validy.dat, which will end up in the Current directory of
% matlab. Attention with each run these files will be overwritten. 
%
% The peak_labelling function is a bit less sensible to image noise, is
% only tracking markers in the image which are actually there and can under
% certain circumstances provide a higher accuracy for larger markers.


[Image,PathImage] = uigetfile('*.tif','Open Image');
cd(PathImage)
load('filenamelist')
filenumber=length(filenamelist);
% filelist_generator

I=imread(Image); % read Image
Itemp=mean(double(I),3);
I=Itemp;
figure, image(I); %show Image
axis('equal');
drawnow
title(sprintf('Mark the region of interest: Click on the on the lower left corner and and then on the upper right corner'))
[xprof, yprof]=ginput(2); % Get the Area of Interest
xmin = xprof(1,1);
xmax = xprof(2,1);
ymin = yprof(2,1);
ymax= yprof(1,1);
tic; % start timer for time estimation
msgboxwicon=msgbox('Subtracting image background, please wait.','Processing...')
I2 = imsubtract (I, imopen(I,strel('disk',15))); % subtract background
close(msgboxwicon)
image(I2); %show with subtracted background
axis('equal');
t(1,1)=toc;
tic;
drawnow
roi=(I2>10); %subtract greyvalues to work only with real peaks
[labeled,numObjects] = bwlabel(roi,8); %label all peaks - very important function, crucial for the whole process; see matlab manual
powderdata=regionprops(labeled,'basic') % get peak properties from bwlabel
powderarea=[powderdata.Area]; %define area variable
powdercentroid=[powderdata.Centroid]; %define position variable
powderboundingbox=[powderdata.BoundingBox]; %define bounding box variable
counter=0;
countermax=length(powdercentroid)/2;
powderxy=zeros(countermax,8);
for i=1:countermax; % get all data from the bwlabel (position, bounding box and area of peaks) 
    counter=counter+1; %
    powderxy(i,1)=i; % number of the detected particle
    powderxy(i,2)=powdercentroid(1, (i*2-1)); % x coordinate of particle position
    powderxy(i,3)=powdercentroid(1, (i*2)); % y coordinate of particle position
    powderxy(i,4)=powderboundingbox(1, (i*4)-3); % x coordinate of bounding box
    powderxy(i,5)=powderboundingbox(1, (i*4)-2); % y coordinate of bounding box
    powderxy(i,6)=powderboundingbox(1, (i*4)-1); % width (x) of bounding box
    powderxy(i,7)=powderboundingbox(1, (i*4)); % height (y) of bounding box
    powderxy(i,8)=powderarea(1, i); % area of bounding box
end


% cropping in x direction to reduce to the area of interest
% crop away peaks which are too small or too big

Amin=10; %minimum peaksize 10 pixel --> only testwise
Amax=1000; %maximum peaksize --> only testwise
counter=0
i=0;

% throw away useless peaks (defined by position and size)
for i=1:countermax; % loop through all points
    
    if xmin<powderxy(i,2) % crop all points left from Region Of Interest (ROI)
        if powderxy(i,2)<xmax % crop all points right from Region Of Interest (ROI)
            if ymin<powderxy(i,3) % crop all points below the Region Of Interest (ROI)
                if powderxy(i,3)<ymax % crop all points above the Region Of Interest (ROI)
                    if Amin<powderxy(i,8) % crop all points with a small peak area 
                        if powderxy(i,8)<Amax % crop all points with a too big area
                            counter=counter+1;
                            cropxy(counter,1)=counter; % peaks get a new number 
                            cropxy(counter,2)=powderxy(i,2); % x
                            cropxy(counter,3)=powderxy(i,3); % y
                            cropxy(counter,4)=powderxy(i,4); % x bounding box
                            cropxy(counter,5)=powderxy(i,5); % y bounding box
                            cropxy(counter,6)=powderxy(i,6); % width (x) bounding box
                            cropxy(counter,7)=powderxy(i,7); % height (y) bounding box
                            cropxy(counter,8)=powderxy(i,8); % area bounding box
                        end
                    end
                end
            end
        end
    end
end

% clear variables
clear powderxy
clear powderdata
clear powderarea
clear powdercentroid
clear powderboundingbox
clear counter
clear countermax
clear Amin
clear Amax
clear roi

close all

t(1,2)=toc; % stop timer


% Start fitting process of the peaks, labeled by bwlabel

tic; % start timer
counter=0
g = waitbar(0,'Processing image'); % nucleating the progress bar
fitcountertemp=size(cropxy); % number off peaks to cycle through
fitcounter=fitcountertemp(1,1); % number off peaks to cycle through
for c=1:fitcounter %start the loop to process all points
    waitbar(c/(fitcounter-1)); % growth of the progress bar
    cropI=imcrop(I,[(round(cropxy(c,2))-round(cropxy(c,6))) (round(cropxy(c,3))-round(cropxy(c,7))) round(cropxy(c,6))*2 round(cropxy(c,7))*2]); % crop the region around the detected peak (bwlabel)
    
% get line profile in x direction for the fitting routine
    
    xdata = [(round(cropxy(c,2))-round(cropxy(c,6))):1:(round(cropxy(c,2))+round(cropxy(c,6)))]; % x-coordinate for the fitting which is equivalent to the x coordinate in the image
    ydata=sum(cropI)/(2*cropxy(c,7)); % y-coordinate for the fitting which is equivalent to the greyvalues in the image - integrated in y direction of the image
    
% fitting in x-direction
    % guess some parameters for the fitting routine --> bad guesses lead to
    % an error message which stops the fitting
    
    back_guess=(ydata(1)+ydata(round(cropxy(c,6))*2))/2; % guess for the background level - averadge of the first and last greyvalue
    sig1_guess=(cropxy(c,6))/5; % guess for the peak width - take a fith of the cropping width
    amp_guess1=ydata(round(cropxy(c,6))); % guess for the amplitude - take the greyvalue at the peak position
    mu1_guess=cropxy(c,2); % guess for the position of the peak - take the position from bwlabel

% start fitting routine
    [x,resnormx,residual,exitflagx,output]  = lsqcurvefit(@gauss_onepk, [amp_guess1 mu1_guess sig1_guess back_guess], xdata, ydata); %give the initial guesses and data to the gauss fitting routine

% show the fitting results
    xtest = [(round(cropxy(c,2))-round(cropxy(c,6))):0.1:(round(cropxy(c,2))+round(cropxy(c,6)))]; % x values for the plot of the fitting result
    ytest = (x(1)*exp((-(xtest-x(2)).^2)./(2.*x(3).^2))) + x(4); % y values of the fitting result
    yguess=(amp_guess1*exp((-(xtest-mu1_guess).^2)./(2.*sig1_guess.^2))) + back_guess; %y values for the guess plot
%     plot(xdata,ydata,'o') % plot the experimental data
%     hold on
%     plot(xtest,ytest,'r') % plot the fitted function
%     plot(xtest,yguess,'b') % plot the guessed function    
%     drawnow
%     hold off
    
% fitting in y-direction
    % guess parameters for the fitting routine --> bad guesses lead to
    % an error message which stops the fitting
   
% get line profile in x direction for the fitting routine
    
    xdata = [(round(cropxy(c,3))-round(cropxy(c,7))):1:(round(cropxy(c,3))+round(cropxy(c,7)))]; % x data in y direction of the image
    ydata=sum(cropI')/(2*cropxy(c,6)); % integrate greyvalues in x direction and normalize it to the number of integrated lines
    
% fitting in y-direction
    % guess parameters for the fitting routine --> bad guesses lead to
    % an error message which stops the fitting
   
    back_guess=(ydata(1)+ydata(round(cropxy(c,7))*2))/2; % guess for the background level - averadge of the first and last greyvalue
    sig1_guess=(cropxy(c,6))/5; % guess for the peak width - take a fith of the cropping width
    amp_guess1=ydata(round(cropxy(c,7))); % guess for the amplitude - take the greyvalue at the peak position
    mu1_guess=cropxy(c,3); % guess for the position of the peak - take the position from bwlabel
    
% start fitting routine
    [y,resnormy,residual,exitflagy,output]  = lsqcurvefit(@gauss_onepk, [amp_guess1 mu1_guess sig1_guess back_guess], xdata, ydata); %give the initial guesses and data to the gauss fitting routine
    
% show the fitting results
    xtest = [(round(cropxy(c,3))-round(cropxy(c,7))):0.1:(round(cropxy(c,3))+round(cropxy(c,7)))]; % x values for the plot of the fitting result
    ytest = (y(1)*exp((-(xtest-y(2)).^2)./(2.*y(3).^2))) + y(4); % y values of the fitting result
    xguess = [(round(cropxy(c,3))-round(cropxy(c,7))):0.1:(round(cropxy(c,3))+round(cropxy(c,7)))]; % x values for the guess
    yguess = (amp_guess1*exp((-(xguess-mu1_guess.^2)./(2.*sig1_guess.^2))) + back_guess); % y values for the guess plpot
%     plot(xdata,ydata,'o') % plot the experimental data
%     hold on
%     plot(xtest,ytest,'g') % plot the fitted function
%     plot(xguess,yguess,'b') % plot the guessed function    
%     drawnow
%     hold off
    
% sort out the bad points and save the good ones in fitxy 
    % this matrix contains the to be used points from the first image
    
    if exitflagx>0 % if the fitting routine didn't find end before the 4000th iteration (check that in lsqcurvefit.m) then exitflag will be equal or smaller then 0
        if exitflagy>0 % the same for the y direction fitting
            if x(3)>1 % the width of the peak should be wider than 1 pixel - this is negotiable: different powder particle or cameras can give back results with very narrow peaks
                if y(3)>1 % the same for y direction fitting
                    if resnormx/cropxy(c,6)<10 % A measure of the "happyness" of a fit is the residual, the difference between the observed and predicted data. (in the help file: Mathematics: Data Analysis and Statistics: Analyzing Residuals)
                        if resnormy/cropxy(c,7)<10 % the same for the y- direction - - - a good value is as far as I know until now between 30 and 50. The good fits stay well beyond that (between 0 and 10)
                            counter=counter+1; 
                            fitxy(counter,1)=c; % points  get their final number 
                            fitxy(counter,2)=abs(x(1)); % fitted amplitude x-direction
                            fitxy(counter,3)=abs(x(2)); % fitted position of the peak x-direction
                            fitxy(counter,4)=abs(x(3)); % fitted peak width in x-direction
                            fitxy(counter,5)=(x(4)); % fitted background in x-direction
                            fitxy(counter,6)=abs(y(1)); % fitted amplitude y-direction
                            fitxy(counter,7)=abs(y(2)); % fitted position of the peak y-direction
                            fitxy(counter,8)=abs(y(3)); % fitted peak width in y-direction
                            fitxy(counter,9)=abs(y(4)); % fitted background in y-direction
                            fitxy(counter,10)=cropxy(c,6); % cropping width in x-direction
                            fitxy(counter,11)=cropxy(c,7); % cropping width in ydirection
                        end
                    end
                end
            end
        end
    end
    
end


close(g) % close progress bar window
t(1,3)=toc; % stop timer
image_time_s=t(1,3); % take time per image
estimated_totaltime_h=image_time_s*filenumber/3600 % calculate estimated time
sum(t);
total_time_h=sum(t)
close all

% plot image with peaks labeled by bwlabel (crosses) and the chosen points
% which are easy to fit with a gaussian distribution (circles)

figure, image(I2); %show Image
title(['Number of selected Images: ', num2str(filenumber), '; Estimated time [h] ', num2str((round(estimated_totaltime_h*10)/10)), ' Crosses are determined peaks, circles are chosen for  the analysis. If you want to run the analysis hit ENTER'])
axis('equal');
hold on;
plot(cropxy(:,2),cropxy(:,3),'+','Color','white') % peaks from bwlabel
plot(fitxy(:,3),fitxy(:,7),'o','Color','white'); % "good" points
drawnow

total_progress=1/filenumber;

pause

close all
fitlength=size(fitxy);
fitcounter=fitlength(1,1)
% again for all images
for m=1:(filenumber-1) % loop through all images
    tic; %start timer
    counter=0;
    I = imread(filenamelist(m,:)); %read image
    Itemp=mean(double(I),3);
    I=Itemp;
    f = waitbar(0,'Working on Image');

    
    % loop number
    for c=1:fitcounter %loop trough all points
        waitbar(c/(fitcounter-1)); %progress bar
        
        % load variables
        pointnumber=fitxy(c,(m-1)*12+1);
        amp_guess_x=fitxy(c,(m-1)*12+2);
        mu_guess_x=fitxy(c,(m-1)*12+3);
        sig_guess_x=fitxy(c,(m-1)*12+4);
        back_guess_x=fitxy(c,(m-1)*12+5);
        amp_guess_y=fitxy(c,(m-1)*12+6);
        mu_guess_y=fitxy(c,(m-1)*12+7);
        sig_guess_y=fitxy(c,(m-1)*12+8);
        back_guess_y=fitxy(c,(m-1)*12+9);
        crop_x=fitxy(c,(m-1)*12+10);
        crop_y=fitxy(c,(m-1)*12+11);
        
        % crop the area around the point to fit
        
        cropI=imcrop(I,[(round(mu_guess_x)-round(crop_x)) (round(mu_guess_y)-round(crop_y)) 2*round(crop_x) 2*round(crop_y)]);
%         cropedI=imcrop(I,[(round(mu_guess_x)-round(crop_x)) (round(mu_guess_y)-round(crop_y)) 2*round(crop_x) 2*round(crop_y)]);
%         cropI=imsubtract (cropedI, imopen(cropedI,strel('disk',15))); % subtract background
        %         imshow(cropI)
        % get line profile in x direction
        xdatax = [(round(mu_guess_x)-round(crop_x)):1:(round(mu_guess_x)+round(crop_x))];
        ydatax=sum(cropI)/(2*(crop_y));
        xguessx = [(round(mu_guess_x)-round(crop_x)):0.1:(round(mu_guess_x)+round(crop_x))];
        yguessx = (amp_guess_x*exp((-(xguessx-mu_guess_x).^2)./(2.*sig_guess_x.^2))) + back_guess_x;
        [x,resnormx,residualx,exitflagx,output]  = lsqcurvefit(@gauss_onepk, [amp_guess_x mu_guess_x sig_guess_x back_guess_x], xdatax, ydatax);
        xtestx = [(round(mu_guess_x)-round(crop_x)):0.1:(round(mu_guess_x)+round(crop_x))];
        ytestx = (x(1)*exp((-(xtestx-x(2)).^2)./(2.*x(3).^2))) + x(4);
%         plot(xdatax,ydatax,'o')
%         hold on
%         plot(xtestx,ytestx,'r')
%         plot(xguessx,yguessx,'b')
%         title(['Filename: ',filenamelist(m,:), '; Progress [%]: ',num2str((round(total_progress*10))/10), '; Tot. t [h] ', num2str((round(total_time_h*10)/10)), '; Est. t [h] ', num2str((round(estimated_totaltime_h*10)/10))])
%         drawnow
%         hold off
        xdatay = [(round(mu_guess_y)-round(crop_y)):1:(round(mu_guess_y)+round(crop_y))];
        ydatay=sum(cropI')/(2*(crop_y));
        xguessy = [(round(mu_guess_y)-round(crop_y)):0.1:(round(mu_guess_y)+round(crop_y))];
        yguessy = (amp_guess_y*exp((-(xguessy-mu_guess_y).^2)./(2.*sig_guess_y.^2))) + back_guess_y;
        [y,resnormy,residualy,exitflagy,output]  = lsqcurvefit(@gauss_onepk, [amp_guess_y mu_guess_y sig_guess_y back_guess_y], xdatay, ydatay);
        xtesty = [(round(mu_guess_y)-round(crop_y)):0.1:(round(mu_guess_y)+round(crop_y))];
        ytesty= (y(1)*exp((-(xtesty-y(2)).^2)./(2.*y(3).^2))) + y(4);
%         plot(xdatay,ydatay,'o')
%         hold on
%         plot(xtesty,ytesty,'g')
%         plot(xguessy,yguessy,'b')
%         title(['Filename: ',filenamelist(m,:), '; Progress [%]: ',num2str((round(total_progress*10))/10), '; Tot. t [h] ', num2str((round(total_time_h*10)/10)), '; Est. t [h] ', num2str((round(estimated_totaltime_h*10)/10))])
%         drawnow
%         hold off

        
        if exitflagx>0
            if exitflagy>0
                counter=counter+1;
                fitxy(counter,m*12+1)=pointnumber;
                fitxy(counter,m*12+2)=abs(x(1));
                fitxy(counter,m*12+3)=abs(x(2));
                fitxy(counter,m*12+4)=abs(x(3));
                fitxy(counter,m*12+5)=abs(x(4));
                fitxy(counter,m*12+6)=abs(y(1));
                fitxy(counter,m*12+7)=abs(y(2));
                fitxy(counter,m*12+8)=abs(y(3));
                fitxy(counter,m*12+9)=abs(y(4));
                fitxy(counter,m*12+10)=crop_x;
                fitxy(counter,m*12+11)=crop_y;
                fitxy(counter,m*12+12)=resnormx;
                
            end
        end
        
        
    end
    
    plot(fitxy(:,m*12+1),fitxy(:,m*12+12),'+');
    title(['Filename: ',filenamelist(m,:), '; Progress [%]: ',num2str((round(total_progress*10))/10), '; Tot. t [h] ', num2str((round(total_time_h*10)/10)), '; Est. t [h] ', num2str((round(estimated_totaltime_h*10)/10))])
    fitcounter=counter;
    drawnow
    time(m)=toc;
    total_time_s=sum(time);
    total_time_h=sum(time)/3600;
    image_time_s=total_time_s/m;
    estimated_totaltime_h=image_time_s*(filenumber)/3600
    progress_percent=total_time_h/estimated_totaltime_h*100;
    total_progress=(m+1)/(filenumber)*100
        close(f);

end  

% save the stuff
save fitxy.dat fitxy -ascii -tabs

[validx,validy]=sortvalidpoints(fitxy);
title(['Processing Images finished!'])

save('validx');
save('validy');

save validx.dat validx -ascii -tabs;
save validy.dat validy -ascii -tabs;