OBTENTION OF AN ABSORBANCE SPECTRA FROM AN IMAGE

I want to process images from light spectra to obtain an absorbance vs wavelength plot. However, the ligth spectra is only a small part of the image, and i do not understand how to set the wavelengths at the pixels they correspond. That is to say, only the light spectra is half of the image (for example), how do i know when wavelengths start to count in the visible spectra.

[DGM]

We don't know what images or other calibration information you have -- or should be expected to have. It's up to you to describe your problem.

[Image Analyst]

Show your image. I have no idea what it is. Is it like a diffraction pattern coming off a grating? Is it a rainbow-like line coming out of a prism? These two cases would be best. Or do you just have images from an RGB sensor, like a digital camera and need to estimate spectrum from the RGB values in the image? This situation would be least accurate. See this paper:

Digital camera-based spectral estimation in open environment

based on imaging condition correction

Jinxing Liang1, Kaida Xiao2

1College of Mathematics and Computer Science, Wuhan Textile University, Wuhan, Hubei, China

2School of Design, University of Leeds, Leeds, United Kingdom

Abstract

Digital camera-based spectral estimation in open

environment is a challenge in current stage. Although some

methods have been proposed in recent years, the methods do not

consider the exposure inconsistency between camera spectral

characterization and spectral estimation applications, that

makes the proposed method cannot for practical applications.

We proposed here a spectral estimation method based on

imaging condition correction of which can deal with the

problem exist in current methods. Using the whiteboard and raw

camera response, the imaging conditions of open environment is

recorded and corrected to the reference imaging conditions, and

the surface spectral of object is estimated using the established

spectral estimation matrix in the reference imaging conditions.

The proposed method in three application models are tested and

compared. The result shows that the adaptive model for imaging

condition correction gives the best spectral estimation accuracy

[Jorge Trapiella Fernández]

The pics look like this, the problem is that obviously the first pixel does not correspond to 0 nm wavelength and i have no clue on how to interpolate to obtain the relation pic-wavelength.

Thank you for your responses

[DGM]

You need to know two relationships:

1. The spatial position within the image versus wavelength
2. The camera's sensitivity versus wavelength

We still don't know how this image was created. Consequently, we don't know if either of these two things are repeatable. The physical staging and camera behavior will need to be repeatable in order to calibrate.

From appearances alone, the fact that the image is so heavily biased toward the primaries tells me that something unhelpful is happening in the acquisition. I could guess that this might possibly mean that:

• There is some sort of color vibrancy filter applied?
• Sensor saturation? Some sort of internal compensation?
• The image didn't come from a grating or prism?

If you're trying to do some sort of rudimentary thing with a camera and a grating, just do a web search for "theremino calibration".

Answer 1

[Star Strider]

Without some sort of calibration, what you want to do might not be possible, especially if the resolution is only integers on the range of .

With that in mind, try this —

I = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1708001/image.jpeg');

SzI = size(I)

SzI = 1x3

102 282 3

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

imshow(I)

Irow = I(51,:,:);
sqIrow = squeeze(Irow);

cc = 'rgb';
dn = ["Red" "Green" "Blue"];

figure
hold on
for k = 1:size(I,3)
plot(sqIrow(:,k), cc(k), 'DisplayName',dn(k)+" Channel")
end
hold off
grid
xlabel('Row Index')
ylabel('Channel Value')
title('Centre Row')
legend('Location','best')

Irow = mean(I,1);
sqIrow = squeeze(Irow);

cc = 'rgb';
dn = ["Red" "Green" "Blue"];

figure
hold on
for k = 1:size(I,3)
plot(sqIrow(:,k), cc(k), 'DisplayName',dn(k)+" Channel")
end
hold off
grid
xlabel('Row Index')
ylabel('Channel Value')
title('Mean Over Rows')
legend('Location','best')

SEM = squeeze(std(double(I),[],1)/sqrt(size(I,1)));
t95 = tinv([0.025 0.975], size(I,2)-1);
for k = 1:size(I,3)
    CI95{k} = sqIrow(:,k) + SEM(:,k) .* t95;
end

cc = 'rgb';
dn = ["Red" "Green" "Blue"];
xv = [(0:size(I,2)-1) flip(0:size(I,2)-1)].';

figure
hold on
for k = 1:size(I,3)
hp(k) = plot(sqIrow(:,k), cc(k), 'DisplayName',dn(k)+" Channel");
yv = [CI95{k}(:,1); flip(CI95{k}(:,2))];
patch(xv, yv, cc(k), 'FaceAlpha',0.2, 'EdgeColor','none')
end
hold off
grid
xlabel('Row Index')
ylabel('Channel Value')
title('Mean Over Rows With 95% Confidence Intervals')
legend(hp, 'Location','best')

EDIT — (2 Jun 2024 at 19:03)

Changed confidence interval plots from dashed line plots to patch objects.

.

[Image Analyst]

Yes but to then turn the 3 channel absorbances into an overall absorbance spectra vs wavements (which is a single plot of absorbance vs. wavelength), you'll need to know the spectral sensitivities of the imaging sensor as a function of wavelength, and what wavelengths are at the ends of the colored image.

[Star Strider]

Certainly. Much information remains missing.

My objective is to demonstrate how to get the information from the spectrum figure, and calculate some basic dispersion statistics on it.