Filtering Blobs on a Binary Image

Hi, I have a 12bit grayscale image of some beads. I want to make measurements on the central one and as a first approach, I want to remove all the other blobs and replace their values with an average of the background

I've identified the most central blob. Is there away to remove the other blobs?

Also, for the area filtering, I have hard coded the limits, but it will change for my images . Im looking for a way to get a ball park estimate of the size of the objects (area). I was thinking about c=normxcor with a gaussian template and loop it about 5 times with different gaussian fwhm's and see which one throws up the highest value of c. Is this a good / bad idea, any other ways to get a ball park figure of the size (area)?

Thanks

        ax=app.UIAxes2; clc;
        IMraw=getimage(ax); IM=IMraw-min(IMraw(:));    %Background 
        [sy,sx]=size(IM);  xc=sx/2; yc=sy/2;

        % Create Binary Image
        mn=mean2(IM);
        thresholdValue = mn; 
        BW = IM > thresholdValue;  % Binary image

        myfigure(app,1200,400);
        tcl = tiledlayout(1,4,'TileSpacing','compact','Padding','compact');   %tight
        ax1=nexttile;   
        myImagesc(app,ax1,IM);  title(ax1,'Raw Image'); 
        
        ax2=nexttile; 
        myImagesc(app,ax2,BW);  title(ax2,'Binarised');

        ax3=nexttile;
        img1 = xor(bwareaopen(BW,40),  bwareaopen(BW,100));   % Area Filter Here    (Can also use bwareafilt here)        
        myImagesc(app,ax3,img1);  title(ax3,'Area Filter');

        % Try To Ientify blob nearest centre of image - need
        % RegionProps

        s = regionprops(BW,'Centroid','Circularity');
        cen = cat(1,s.Centroid);  circ=cat(1,s.Circularity);
        x1=cen(:,1); y1=cen(:,2);

        % Find Distance of Blob centroids from centre of Image
        n=numel(x1);
        data=[];
        if n>0   
            for i=1:n
                X = [x1(i),y1(i);xc,yc];
                d = pdist(X,'euclidean');
                Intensity=IM(round(y1(i)),round(x1(i)));
                data(i,1)=1; data(i,2)=x1(i); data(i,3)=y1(i); data(i,4)=d; data(i,5)=Intensity;  data(i,6)=circ(i,1);
            end
        end    

        % Sort rows in Distance D from xc,yc
        B= sortrows(data , 4,"ascend")

        % Plot on Binary the one to keep
        xkeep=B(1,2); ykeep=B(1,3); hold(ax3,'on');
        plot(ax3,xkeep,ykeep,'r*');

[Jason]

I've just found this seems to work for isolating the central spot

BW2 = bwselect(BW,xkeep,ykeep);

But the subtraction doesn't work

%Subtract from BW
mask=BW-B;

[Jason]

So I've got a little further and dilated the central blob abit.

Couple of follow on queastions

1: How to make the "Keep" Blob exactly circular - that then becomes my mask?

2: I've currently put every ~mask as the median of the min 10 pixels, but its too low. Any suggestions to getting the median value of the pixels in the perimeter of the mask?

       % Sort rows in Distance D from xc,yc
        B= sortrows(data , 4,"ascend");
        B

        % Plot on Binary the one to keep
        xkeep=B(1,2); ykeep=B(1,3); hold(ax3,'on');
        plot(ax3,xkeep,ykeep,'r*');

        BW2 = bwselect(BW,xkeep,ykeep);     % Removes all other blobs
        SE2 = strel('disk',6);              % would like to "circularise and expand a bit"
        BW3 = imdilate(BW2,SE2);
        ax4=nexttile; 
        myImagesc(app,ax4,BW3);  title(ax4,'Keep');

        %Apply Mask
        mask=BW3;
        median(mink(IM(:),10))
        IM(~mask)=median(mink(IM(:),10));
        ax5=nexttile; 
        myImagesc(app,ax5,IM);  title(ax5,'Mask');

Answer 1

[Matt J]

If you can form an under-estimate of the extent of the blobs, let's call them sub-blobs, then you can fit a Gaussian lobe to the sub-blob pixels, using for example, this FEX download:

https://www.mathworks.com/matlabcentral/fileexchange/69116-gaussfitn

The fitted covariance matrix will then give you a sharper estimate of the full blob size.

[Matt J]

1: How to make the "Keep" Blob exactly circular - that then becomes my mask?

Once you've estimated the center and radius of the mask, you can circularize it with

h=drawcircle('Center',___,'Radius',__);
mask=h.createMask;
delete(h);

or

V=nsidedpoly(1000,'Center',___,'Radius',____).Vertices;

mask=poly2mask(V(:,1),V(:,2),sy,sx);

2: I've currently put every ~mak as the mdeain of the min 10 pixels, but its too low. Any suggestions to getting the median value of the pixels in the perimeter of the mask?

If you use gaussfitn, as I suggesed above, then the estimated D parameter will be an estimate of the background.