Re: histogram matching of colour images

[Terry Duell]

Hello Andreas,

On Mon, 31 Aug 2015 16:53:54 +1000, Andreas Weber <address@hidden>  
wrote:

> Am 31.08.2015 um 04:33 schrieb Terry Duell:
> > Hello All,
> > I am looking for a method of matching the histogram of a subject image
> > to that of a reference image, both r,g,b.
> > I don't see any functions in the image package that appear to do that,
> > but may be suffering domestic blindness.
> > Anyone have any ideas how to do this?
>
> I don't know a function in octave-forge image which would do this but it
> should be easy to do it yourself.
>
> ## get octave logo
> img = ind2rgb (get (0, "defaultimagecdata"), colormap);
>
> bins = 10;
> n = zeros (3, bins);
> n(1,:) = hist (img (:, :, 1)(:), bins);
> n(2,:) = hist (img (:, :, 2)(:), bins);
> n(3,:) = hist (img (:, :, 3)(:), bins);
>
>
> Then you'll have the counts of red, green, blue pixels per bin which I
> would normalize to 0..1 and then feed to a "cost" function, perhaps
>
> sumsq (n_ref - n)
>
> to calculate how "similar" the two images are.
> HTH, Andy

I only just found this function, which looks like it might do what I  
want...not really tested it yet.

function I = histmatch(I1, I2)
%% Histogram Matching from image I1 to image I2
% I1: reference image
% I2: target image

cdf1 = cal_cdf(I1);  % cdf of image I1
cdf2 = cal_cdf(I2);  % cdf of image I2

%% mapping function Fk()
Fk = zeros(1, 256);
for k = 1:256
    tt = abs(cdf2(k) - cdf1);
    [~, idx] = min(tt);
    Fk(k) = idx - 1;
end

% mapping
[r, c] = size(I2);
I2 = I2(:);
I = I2;
for k = 0:255
    I(I2 == k) = Fk(k+1);
end
I = reshape(I, r, c);

end


function t_cdf = cal_cdf(I)
%% calculate cumulative distribution function (CDF) of image I
[hist, ~] = imhist(I);
hist = hist / numel(I);  % normalize hist
t_cdf = cumsum(hist);    % cdf
end

Thanks for your help.

Cheers,
--
Regards,
Terry Duell