How to align two images together automatically: Round 2

Hello!

I have my images closely aligned, but not close enough. Is there anyway I can improve object selection in an image or getting the centroid more accurately/ consistently? I have been using this script to locate the centroid of both images and Sean De's imtranslate to modify the image.

Thanks

-Frank

Image

Code

%First Image
i = imread('343TESTA.tif');
j = size(i);
imtool(i);

for k = 1:j(1)
    for l = 1:j(2)
        if i(k,l) > 160 & i(k,l) > 160
            d(k,l) = 1;
        else
       d(k,l) = 0;
        end
    end
end
imtool(d);

labela = bwlabel(d);
imagesc(labela);

[labela,num] = bwlabel(d,4);

stats = regionprops(labela,'basic');

stats(1).Area
stats(1).Centroid
max_area = max([stats.Area]);
biggrain = find([stats.Area]== max_area);
p = stats(biggrain).Centroid
%Second Image
a = imread('343TESTB.tif');
b = size(a);
imtool(a);

for c = 1:b(1)
    for l = 1:b(2)
        if i(c,l) > 160 & a(c,l) > 160
            e(c,l) = 1;
        else
       e(c,l) = 0;
        end
    end
end
imtool(e);

labelb = bwlabel(e);
imagesc(labelb);

[labelb,num] = bwlabel(e,4);

stats = regionprops(labelb,'basic');

stats(1).Area
stats(1).Centroid
max_area = max([stats.Area]);
biggrain = find([stats.Area]== max_area);
o = stats(biggrain).Centroid
%Calculates the differences in the centroid

displacement = p -o
%IM Translate

v = imread('343TESTA.tif');
u = imread('343TESTB.tif');
u2 = imtranslate(u, [0 -1.0376]); %Note: [y x] for imtranslate
subplot(1,2,1), imshow(v)
subplot(1,2,2), imshow(u2)
imwrite(u2, '343TESTC','tif');

[Frank]

I'm using Fiji (ImageJ variant) to see how this script aligns the images. Stackreg isn't good because it moves the image up and down, I only want left and right alignment.

Answer 1

[Sean de Wolski]

http://www.mathworks.com/matlabcentral/fileexchange/18401-efficient-subpixel-image-registration-by-cross-correlation

You could try this, though I don't know how well it will work on binary images...

Also, how were these images acquired? Is there a chance that there is a change in thickness/density etc.?

[Frank]

Not likely, these images are probably the best acquisition of the subject that technology can get today.

Would brightness of the image help?

[Sean de Wolski]

Maybe. Perhaps try the subpixel phase correlation on the original images, not the images converted to binary.