Background color correction of a rgb picture

Recently i took this picture with a thermocamera, the color of the background should be homogeneus and more or less similir to the one in the center (red/orange), but on the corners of the picture due to small field of view of the camera it is violet/blue. how i can make it more uniform?? thanks in advance

Answer 1

[Image Analyst]

Try this:

% Demo by Image Analyst
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear all;
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 16;

%===============================================================================
% Get the name of the image the user wants to use.
baseFileName = 'prova2.jpg';
folder = pwd;
fullFileName = fullfile(folder, baseFileName);
% Check if file exists.
if ~exist(fullFileName, 'file')
    % The file doesn't exist -- didn't find it there in that folder.
    % Check the entire search path (other folders) for the file by stripping off the folder.
    fullFileNameOnSearchPath = baseFileName; % No path this time.
    if ~exist(fullFileNameOnSearchPath, 'file')
        % Still didn't find it.  Alert user.
        errorMessage = sprintf('Error: %s does not exist in the search path folders.', fullFileName);
        uiwait(warndlg(errorMessage));
        return;
    end
end

%=======================================================================================
% Read in image.
rgbImage = imread(fullFileName);
% Get the dimensions of the image.
[rows, columns, numberOfColorChannels] = size(rgbImage)
% Display image.
subplot(3, 1, 1);
imshow(rgbImage, []);
impixelinfo;
axis on;
caption = sprintf('Original RGB Image\n%s', baseFileName);
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo(); % Set up status line to see values when you mouse over the image.
% Set up figure properties:
% Enlarge figure to full screen.
g2 = gcf;
g2.WindowState = "maximized";
% Get rid of tool bar and pulldown menus that are along top of figure.
% set(gcf, 'Toolbar', 'none', 'Menu', 'none');
% Give a name to the title bar.
set(gcf, 'Name', 'Demo by ImageAnalyst', 'NumberTitle', 'Off')
drawnow;

%=======================================================================================
% Read in template/mask image.
fullFileName = fullfile(folder, 'prova2 mask.png');
maskImage = imread(fullFileName);
% Get the dimensions of the image.
[rows, columns, numberOfColorChannels] = size(maskImage);
% Make sure it's binary.
if numberOfColorChannels > 3
    maskImage = maskImage(:, :, 1) > 0;
else
    maskImage = maskImage > 0;
end
% Display image.
subplot(3, 1, 2);
imshow(maskImage, []);
impixelinfo;
axis on;
caption = sprintf('Mask Image');
title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo(); % Set up status line to see values when you mouse over the image.

%--------------------------------------------------------------------------------------------------
% Get the bottom part of the image
midRow = floor(rows / 2);
rgbImage2 = rgbImage(midRow:end, :, :);
mask2 = maskImage(midRow:end, :);
% Split up into individual color channels.
[r2, g2, b2] = imsplit(rgbImage2);

%--------------------------------------------------------------------------------------------------
% Get the mean RGB of the bottom half of the image outside of the mask
meanR = mean(r2(mask2));
meanG = mean(g2(mask2));
meanB = mean(b2(mask2));

%--------------------------------------------------------------------------------------------------
% Split up into individual color channels.
[r, g, b] = imsplit(rgbImage);
% Make everywhere outside the mask that color.
r(~maskImage) = meanR;
g(~maskImage) = meanG;
b(~maskImage) = meanB;
% Reconstruct RGB image.
rgbImage3 = cat(3, r, g, b);
% Display image.
subplot(3, 1, 3);
imshow(rgbImage3, []);
impixelinfo;
axis on;
title('Background-replaced Image', 'FontSize', fontSize, 'Interpreter', 'None');
hp = impixelinfo(); % Set up status line to see values when you mouse over the image.
drawnow;

[Gianmaria Gasperoni]

No this in not what i want to obtain. I cannot use a template for the background. the information related to the background are fundamental and doing like this i lose them all. in the comment on the top where we were speeking about this https://it.mathworks.com/help/images/correcting-nonuniform-illumination.html. you told that u applied the same logic to rgb picture.how you have done that?

[Image Analyst]

I'm not sure what you want now. You said "the color of the background should be homogeneus and more or less similir to the one in the center (red/orange), but on the corners of the picture due to small field of view of the camera it is violet/blue. how i can make it more uniform??" So I have a mask that separates the foreground (squares) from the background (non-squares). The background is not uniform, as you said. I took the average color of the uniform part of the background in the lower half of the image, and I applied that to the entire background so that now the entire background is uniform. I thought that is what you wanted when you asked "how i can make it more uniform" Evidently not, and you want to background to look different somehow. How do you want it to look? Can you use Photoshop or something to mock up a picture of what you think the desired uniform-background image would look like?

[Gianmaria Gasperoni]

Yes understood. the problem is that i dont have a target to obtain, beacuse who also assign to me this work doesen't know how to do that. i'll try to be more clear beacuse i done a lot of mess.

saying homogeneus i dont mean equal,becuase otherwise i'll loose all the information that obtainable. my target is to have the color on the top part, or where i have the violet/blue similar or of closer intensity to the one in the middle/bottom. for example(i'm assuming random values) if the red is 70 and the violet is 150, at the end of the work i could be happy if the color achieved on the top is 95 a trade of between the two. (closer intensity). the target is to obtain this without changing the information on the rest of the picture.

and i dont want to do it brutally maybe using an algoritm to smooth the color beacuse i want to do the same also with other colormap.

i hope to be more clear now.

thanks for your patience

[Image Analyst]

Still not sure. Do you want to concentrate just on the background and fit a smooth surface to the background pixels so you can divide that out and "flatten" the background while leaving all of the little noise, defects, and structure of the background? Do you want the image flattening to affect the squares also, or just the background?

I'm attaching a background correction demo.

[Gianmaria Gasperoni]

It's enough to focus on the background. because then i will use a mask to copy back the squares on their position.

[Image Analyst]

OK but what do you want to do with it? Background flattening like I gave you demos for?

Why would you want to flatten the background ONLY? That doesn't make sense to me and I'd need some kind of justification or rationalization for why that is appropriate. And are you going to analyze both background and foreground separately? Or just one or the other?

Attached is another demo that uses scatteredInterpolant to generate values over the entire image even though you have values for only part of the image (the background which has "holes" where the squares are).

[Gianmaria Gasperoni]

The result that i obtain using your demo is ok! is possible to obtain that with an RGB picture?

I want to flatten only the background because on the square the effect of the noise is not so high and can be considered negligible. while on the background is quite high.then i'll analyze in this order: Only the square, then only some portion of the back and then the square sorrounded by part of the background.

[Image Analyst]

If you flatten the background then of course the values and noise will change so your mean and stdev will be that of your "repaired" background rather than what you actually had. The mean will of course average out much of the pixel-to-pixel noise. There are other denoising methods you could use if you want, though like I said, they probably won't change the mean much. And since the background was flattened but the squares weren't you can't compare background levels to square levels.

[Gianmaria Gasperoni]

Understood you are right. in the meanwhile, that i'm going speek with my tutor, can you tell me how i can do the flattening of a RGB picture?

Thanks

[DGM]

You have to do the flattening on the grayscale version.

Using imflatfield() directly on the RGB image flattens V in HSV. The resulting image no longer works with the original colormap. All scale information is lost. It's also worth noting that since V is a poor measure of perceived lightness, the operation doesn't really accomplish much perceived flattening.

% the false-color image
inpict = imread('prova2.jpg');

% flatten the image by flattening V in HSV
flattenedhsv = imflatfield(inpict,10);
imshow(flattenedhsv)

You could flatten each RGB channel, but again that results in an image which does not adhere to the colormap. Also, stretching B emphasizes a bunch of compression artifacts that used to be buried near black.

% flatten the image in RGB
flattenedrgb = inpict;
for c = 1:3
    flattenedrgb(:,:,c) = imflatfield(inpict(:,:,c),10);
end
imshow(flattenedrgb)

Convert to grayscale with the appropriate color map, do the operations, and then convert back (if necessary). The result is an image which can still be read using the same map, though it's questionable whether the particular temperature values correspond to the surface temperature anymore.

% do the flattening on the grayscale copy as before
CT = ccmap_short('flir1',256);
Trange = [30 80]; % complete guess

tpict = rgb2ind(inpict,CT,'nodither');
tpict = rescale(tpict,Trange(1),Trange(2));
flattened = imflatfield(tpict,10);

flattened = mat2gray(flattened,Trange); % normalize
flattened = gray2ind(flattened,size(CT,1)); % quantize
flattened = ind2rgb(flattened,CT);

imshow(flattened)

Answer 2

[Praveen Reddy]

Hi Gianmaria,

I understand that you wanted to have your image background a uniform color. The following resources might help you:

• matlab code to make a background of color image a uniform background
• change the black background to blue same as the image color

[Gianmaria Gasperoni]

First of all thanks for the answer. Maybe i didn't explain it well. So the problem in not on the squares, the problem is on the background. The color of the background in the center and bottom part of the picture is the real one and i want to obtain somenthing similar on the top part where the blue and violet are dominant(without changing the palette of color).

i hope i was clear and thanks in advance

Answer 3

[Image Analyst]

See attached demo. Adapt as needed. You will need to come up with a mask for the squares. Hopefully they are in the same location for all images and you can just make up a template manually.

[Gianmaria Gasperoni]

First of all thanks for the answer. Maybe i didn't explain it well. So the problem in not on the squares, the problem is on the background. The idea of using a mask for the square is ok, but after "removing" them from the picture i need to work on the background. and is this the problem. The color of the background in the center and bottom part of the picture is the real one and i want to obtain somenthing similar on the top part where the blue and violet are dominant(without changing the palette of color).

i hope i was clear and thanks in advance

[Image Analyst]

Yes, I knew all that. So you didn't answer my questions about whether you can use a template, like if your camera is securely mounted to some sort of frame or rig. You could get the mean color from the bottom half of the picture and then apply that color to the whole template.

[Gianmaria Gasperoni]

Yes u are right,sorry.

Im not sure that i can use a template, i think not but i will ask.

My teacher told me to take inspiration from this https://it.mathworks.com/help/images/correcting-nonuniform-illumination.html . i have understood the logic behind but applying the same procedure on a rgb picture the result obtained is absolutely not good.

[Image Analyst]

Well I did it on an RGB image. The logic is about the same.

Answer 4

[DGM]

This looks more like a slightly off-normal camera position and a reflective surface.

Either way, you don't operate on the false-color image. You convert the image to a grayscale representation of temperature. For that, you'd need to know the colormap and the scaling.

% the false-color image
inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1315795/prova2.jpg');

% i only assume that this is approximately the map in use
CT = flirmap('flir1',256);

% i'm going to assume that the mapping is linear
% and that these are the extents of T represented
Trange = [30 80]; % complete guess

% convert from false-color to temperature
% this image will be quantized
tpict = rgb2ind(inpict,CT,'nodither');
tpict = rescale(tpict,Trange(1),Trange(2));

hi = imshow(tpict,Trange);
datatip(hi,160,120,'location','northwest');
datatip(hi,670,140,'location','northeast');
datatip(hi,100,600,'location','southeast');

At this point, you can use imflatfield() or whater you want on it, but you're going to be left with the problem of losing your temperature information. You'll be left with a new image whose scale has nothing to do with the original. It's anybody's guess what the actual temperatures are now.

% flatten the image
flattened = imflatfield(tpict,10);
hi = imshow(flattened,Trange);
datatip(hi,160,120,'location','northwest');
datatip(hi,670,140,'location','northeast');
datatip(hi,100,600,'location','southeast');

If the temperature of the background is what matters, then you've lost that information. It's just been roughly averaged out. If you can assume that it's some uniform temperature, then you'd have to offset the result accordingly.

Of course, the temperature of the squares has been lost and even the local uniformity has been lost due to edge effects, so rescaling based on another assumed temperature reference isn't much of a fix.

Could you make a composite image out of the two? Sure, but that sounds an awful lot like fabricating measurements.

[Gianmaria Gasperoni]

First of all thanks for your answer. The first picture that is showed before(not the flattened) is what i want to obtain, but not in grayscale. SInce i need to have the temperature infomation related to the background (using a mask i can save the T inforamtion of the squares). but i need somenthing to tune the color obtaining a smooth transition. i hope to be clear

and thanks in advance

[DGM]

The first image is the same as the false-color image. It represents the same things. It has the same local variations in the represented values. Nothing has been changed. It can be converted back into the original image.

Changing the color of the false-color image is the same as changing the gray level in the gray image, which is the same as changing the temperatures that are represented.

If what's important is the temperature of the squares, then why does changing the background matter? If the variation in the apparent background temperature is caused by the surface being reflective, are the squares equally reflective and subject to the same error?

[Gianmaria Gasperoni]

First of all the picture on the top is part of a video that i need to analyze. Information regarding the background and the square are both important because allow us to predict some strange behaviur during the process (porosity, cracks etc). the interesting information can be extracted using the T, intensity of each pixel ecc. the squares should be equally reflective because are made by the same material (without considering little variation due to porosity ecc)

[DGM]

I don't know if this is any help. It's more or less a middleground between what IA and I have posted.

% the false-color image
inpict = imread('prova2.jpg');

% i only assume that this is approximately the map in use
CT = flirmap('flir1',256);

% i'm going to assume that the mapping is linear
% and that these are the extents of T represented
Trange = [30 80]; % complete guess

% convert from false-color to temperature
% this image will be quantized
tpict = rgb2ind(inpict,CT,'nodither');
tpict = rescale(tpict,Trange(1),Trange(2));

% flatten the image
flattened = imflatfield(tpict,10);
imshow(flattened,Trange);

% copy the squares back?
mask = im2double(imread('softmask.png'));
outpict = mask.*tpict + (1-mask).*flattened;

imshow(outpict,[])

% convert back to pseudocolor if desired
outpict = mat2gray(outpict,Trange); % normalize
outpict = gray2ind(outpict,size(CT,1)); % quantize
outpict = ind2rgb(outpict,CT);

imshow(outpict)

I don't know if it matters, but I used a soft mask for the composition to avoid creating any false edge features which might interfere with crack detection.

Answer 5

[Image Analyst]

If you want to convert a pseudocolored thermal image to an image where the pixel values are the actual temperatures, see attached demo.

[DGM]

FWIW, the colormap generator I attached above will also generate an approximation of this particular map.

Avoiding colorbar extraction is a big deal, considering that most snapshots are pretty low-quality. They're usually horrible JPGs, so the extracted colormap will always have a curl at the ends.

names = {'flir1','flir2','dias1'};

m = 64; % block height
n = 256; % block width (CT length)
A = repmat(1:n,m,1); % basis

% generate test sweeps
nf = numel(names);
outpict = zeros(m,n,3,nf);
for f = 1:nf
	outpict(:,:,:,f) = ind2rgb(A,flirmap(names{f},n));
end
outpict = imtile(outpict,[nf 1]);

imshow(outpict)

I know there are other maps used by both vendors, but these are the only ones for which I had enough examples to work from.