Is there a way I can convert image (pixel) values ​​to Raman shift using MATLAB?

Hi, I have a little newbie problem using Raman Spectroscopy information. I have a .spc file which by mistake of mine was saved with pixel values for x axis and needed the values to be in Raman shift. Is there a way to convert these image values to Raman Shift for x axis with MATLAB? For the y axis it is the same parameter for both.

[David Goodmanson]

Hi Carlos, could you explain a bit more about what these files represent, and their relationship?

[caogqwerty123]

Hi. The .txt files that I uploaded are two graphs with different parameters on the x axis, what I want to do is that the graph of Calcite, in Pixel values on the x axis, will be the same as the graph of Calcite with values of Raman Shift on the x axis, that is, I want to perform a conversion of units of values from Pixel to Raman Shift, by means of MATLAB. There are scientific articles that mention this conversion through MATLAB but do not say how, only that they used the software...

Answer 1

[Star Strider]

They appear to me to be different data.

However, filtering out the high-frequency noise is likely as close as it is possible to get to correcting the ‘Pixel’ data —

T1 = readtable('https://www.mathworks.com/matlabcentral/answers/uploaded_files/833340/Calcite%20Pixel.txt', 'VariableNamingRule','preserve')

T1 = 1920×2 table
    Pixel    Dark Subtracted
    _____    _______________

      86           2322     
      87         2418.7     
      88         2495.7     
      89         2612.7     
      90         2595.7     
      91         2477.3     
      92         2252.3     
      93         2153.7     
      94         2163.7     
      95         2126.7     
      96         2140.7     
      97           2086     
      98         2080.3     
      99         2049.3     
     100         2049.3     
     101         2046.7     

T2 = readtable('https://www.mathworks.com/matlabcentral/answers/uploaded_files/833345/Calcite%20Raman%20Shift.txt', 'VariableNamingRule','preserve')

T2 = 2629×2 table
    Raman Shift    Dark Subtracted
    ___________    _______________

        162            1512.2     
        163            1431.5     
        164            1366.8     
        165            1319.7     
        166            1271.8     
        167            1228.4     
        168            1172.3     
        169            1126.9     
        170            1116.9     
        171            1139.4     
        172            1139.1     
        173              1124     
        174            1106.7     
        175            1090.3     
        176            1078.9     
        177            1070.1     

Pixx = T1.Pixel;
PixDS = T1.('Dark Subtracted');

RSx = T2.('Raman Shift');
RSDS = T2.('Dark Subtracted');

figure
plot(Pixx, PixDS)
grid
title('Pixel')

figure
plot(RSx, RSDS)
grid
title('Raman Shift')

PixDScb = correctBaseline(Pixx, PixDS, 5, 50, 3);

figure
plot(Pixx, PixDScb)
grid
title('Pixel — Corrected Baseline')

RSDScb = correctBaseline(RSx, RSDS, .01, 5);

Warning: Polynomial is badly conditioned. Add points with distinct X values, reduce the degree of the polynomial, or try centering and scaling as described in HELP POLYFIT.

figure
plot(RSx, RSDScb)
grid
title('Raman Shift — Corrected Baseline')

Pixx_stats = [mean(diff(Pixx)); std(diff(Pixx))];
Fs = 1;
Fn = Fs/2;
L = numel(Pixx);
NFFT = 2^nextpow2(L);
FTPixDS = fft(PixDScb,NFFT)/L;
Fv = linspace(0, 1, NFFT/2+1)*Fn;
Iv = 1:numel(Fv);

figure
plot(Fv, abs(FTPixDS(Iv))*2)
grid
xlim([0 0.1])

Passband = 0.03;                                    % Hz (Or Cycles/(Whatever The x-Axis Units Are))
PixDScb_f = lowpass(PixDScb, 0.03, Fs);

figure
subplot(2,1,1)
plot(Pixx, PixDScb)
grid
title('Pixel — Corrected Baseline')
subplot(2,1,2)
plot(Pixx, PixDScb_f)
grid
title('Pixel — Corrected Baseline — Filtered')

function correctedSignal = correctBaseline(varargin)
% % % ARGUMENT LIST (IN ORDER) — 
% % %   x:          X-Vector
% % %   y:          Y=Vector
% % %   prom:       Peak Minimum Prominence
% % %   sep:        Miinimum Peak Separation
% % %   pd:         Polynomial Degree
% % % 
% % % CALLING CONVENTION — 
% % % correctedSignal = correctedBaseline(x, y, prom, sep, pd_

% vn = numel(varargin)
x = varargin{1};
y = varargin{2};
prom = varargin{3};
sep = varargin{4};
pd = 4;
if numel(varargin) > 4
    pd = varargin{5};
end

TF = islocalmin(y, 'MinProminence',prom, 'MinSeparation',sep);

p = polyfit(x(TF), y(TF),4);
BL = polyval(p, x);
correctedSignal = y-BL;

end

.

[caogqwerty123]

So there is no way that the graph with Pixel values could be the same as with Raman Shift values? Thanks for the reply...

[Star Strider]

I am not certain, since I have no idea how they are related (for example how the pixel values were derived from the Raman data, or the other way round).

So there could be a way, however it is not obvious to me, and apparently not by using the sort of signal processing techniques I used here.

Significantly more detail on the process that created one from the other (whatever that may be), would be helpful, although I definitely cannot guarantee results in any event..

.

[caogqwerty123]

Sorry, it's my fault for not being specific, I really appreciate your effort to understand and the patience that goes with it. Let me explain the relationship... Both graphs represent characteristic Raman signals, that is, some characteristic peaks, in this case, are from Calcite. The Raman Shift graph is the reference to identify that a sample has calcite, how is it identified? Locating the characteristic peaks that are 1092, 715, 285 in Raman Shift. So what I want to achieve is that, in the graph with Pixel values, you can translate the three most prominent peaks of the graph to the same values for Raman Shift, in order to have a way to calibrate graphs with Pixel values in Raman Shift by means of Calcite. If the graphs are superimposed, it is seen that the characteristic peaks in Pixel are further behind in x values than in Raman Shift.

[David Goodmanson]

Hi ortizgc,

the blue and red plots don't look much like each other, so could you comment on which peaks are supposed to match up?

[caogqwerty123]

Hello, the points to be related are:

For Raman Shift: 280.05051, 1085.32579, 711.46459. What are the values to identify that the sample contains Calcite.

For Pixel: Using Origin 2021 and my perception, I was able to identify what could be the characteristic peaks for Calcite being in the Pixel range: 155.56834, 624.69582, 396.54987.

I clarify again, what I want to achieve is that the peaks positioned at 155.56834, 624.69582, 396.54987 in Pixel range, are positioned at 280.05051, 1085.32579, 711.46459, respectively so that I can mention them in Raman Shift instead of Pixel.

[Image Analyst]

Does pinning each end of the pixel signal to the actual, known wavenumber, like I showed below in my Answer, not work? If the signals were looking at a material with the same stuff in each sample, and the wavenumbers were the same for each sample, then it should work.

How did you end up with only pixel values, instead of the actual intensity, anyway? Why don't you have the actual original floating point data from the Raman instrument? Why did the original spc file get overwritten with pixel data? Can you get the original SPC file back from the instrument?

[caogqwerty123]

Pixel signal to the known real wave number because the Raman Shift graph moves when doing that, that is, the characteristic peaks move in the graph that you made, and the characteristic peaks that I observe in the Pixel graph, do not remain in the position they should in Raman Shift. What I really want is that the Pixel graph can be converted to Raman Shift in such a way that the converted Pixel graph remains the same as the Raman Shift graph.

I ended up only with the Pixel values ​​because in the equipment I used, I did not notice that, when saving in .spc, it only saves in an X-axis value, that is, I only saved in Pixel, and not in Raman Shift... And yes, I have the original .spc file. I realized 2 days after doing the sample measurements that we only had the opportunity to measure at that time, in fact, I have more Pixel graphs to convert, the problem is that I do not know what they are because there is no database for Graphs with X axis of Pixel for Raman analysis.

Thank you very much for your answer.

[Image Analyst]

Now I'm confused even more. You talk like "Pixel" is some specific program. Is it? I assumed that somehow you'd gotten coordinates by scanning some image of a plot/graph and the coordinates were in pixels.

If you have the original .spc files then why mess with anything else? Just use that.

[caogqwerty123]

No, Pixel, it is a value of for the X axis, as is Raman Shift. In the software that I use to make the measurements, you can establish in what units of the X axis you want, the problem is that, when saving in .spc, it was saved in Pixel values and not Raman Shift and I cannot revert it because when open it in the software, it won't let me...

[Image Analyst]

Again, I can do it if you just give me the starting and ending wavenumber for each signal.

[caogqwerty123]

For Raman Shift the values for the x-axis are 280.05051, 1085.32579, 711.46459. What are the values to identify that the sample contains Calcite. For Pixel the values for the x-axis could be 155.56834, 624.69582, 396.54987.

[Image Analyst]

OK, so for one signal (which one), it starts at 280 cm-1, and ends at 1085 cm-1, and I ignore the 711.

And for the other signal, it starts at what and ends at what?

I don't need the pixel values because you said those are in the first column of the text file, so we know those. Calcite's x (pixel value) goes from 86 to 2005, and Calcite Raman Shift's x (pixel value) goes from 162 to 2790 -- these are just values from the first column of the text files and I presume these are the pixel coordinates.

OK, let's try a different approach. Are you saying that one signal is a scaled and shifted version of the other and that we need to rescale and shift one signal such that certain spikes align with the corresponding peaks on the other signal? Like maybe you want to do normalized cross correlation? Or just find the two tallest peaks in each signal and align those?

Answer 2

[Image Analyst]

Assuming that both signals were acquired with the same wavenumber range, you can simply rescale the x axis for each to go linearly from the starting wavenumber to the ending wavenumber. However since each signal for some reason has a different number of samples, you'll have to interpolat. See my code below. I'm assuming that the first element of each signal should be relocated to wavenumber1 and the last element of each signal should be relocated to wavenumber2. If we assume that they look reasonably well aligned now, and we have a new x axis that is the wavenumbers. Adapt my code below to use the actual wavenumbers for the signals.

% Read in first signal.
data1 = readmatrix('Calcite Pixel.txt');
x1 = data1(:, 1); % Get pixels.
signal1 = data1(:, 2); % Get intensity.
% Read in other signal
data2 = readmatrix('Calcite Raman Shift.txt');
x2 = data2(:, 1); % Get pixels.
signal2 = data2(:, 2);
subplot(2, 1, 1);
plot(x1, signal1, 'r-', 'LineWidth', 2);
hold on;
plot(x2, signal2, 'b-', 'LineWidth', 2);
grid on;
ylabel('Signal')
xlabel('Pixels')
wavenumber1 = 1700; % Whatever
wavenumber2 = 4000; % Whatever
% Find the max number of wavenumber samples
numSamples = max([length(signal1), length(signal2)])
% Get wavenumbers going from wavenumber1 to wavenumber2 with numSamples
wavenumbers = linspace(wavenumber1, wavenumber2, 2*numSamples);
% Interpolate both signals to be on a common x axis.
xi1 = linspace(wavenumber1, wavenumber2, length(signal1));
signal1 = interp1(xi1, signal1, wavenumbers);
xi2 = linspace(wavenumber1, wavenumber2, length(signal2));
signal2 = interp1(xi2, signal2, wavenumbers);
subplot(2, 1, 2);
plot(wavenumbers, signal1, 'r-', 'LineWidth', 2);
hold on;
plot(wavenumbers, signal2, 'b-', 'LineWidth', 2);
grid on;
ylabel('Signal')
xlabel('Wavenumber')

The top is the original, pixel-based signals, and the bottom is the new signals shfted, interpolated, and rescaled so that the x axis now represents wavenumbers.

[caogqwerty123]

The two graphs are in different range, and what I want to do is that I can convert Pixel values to Raman Shift (they are units of different range), through the characteristic peaks of Calcite in both ranges.155.56834, 624.69582, 396.54987 in Pixel range, and 280.05051, 1085.32579, 711.46459 in Raman Shift.

[Image Analyst]

Why does a range have 3 numbers instead of 2 for the max and min along the x axis?

If you have different starting and stopping wavenumber pairs for each signal, my code can be modified. If you can't do it, let me know what the wavenumber range pair is for each signal and I'll do it for you.

I'm sure you noticed that I didn't do background correction. I use a different method (convex hull) for our Raman signals than Star used (4th order polynomial). I know there are lots of methods. What method do you normally use?

[caogqwerty123]

The range has 3 numbers instead of 2 because the third is one that is already registered in Raman Shift, and more or less I located it in Pixel knowing how to interpret Raman signals.

Could you make the relation of the 3 characteristic peaks of Pixel and Raman Shift to make a kind of calibration and to be able to do the graph conversion? I'm still learning how to use MATLAB, I'm very new to this, sorry...

The background correction method that I used was testing, it was the one that convinced me the most, I have yet to see which one I should use as the signals can be seen, but if there is a method that you could recommend, I will gladly use it.

Thanks for the answer and the patience of understanding.

[Image Analyst]

Now I'm confused even more. You talk like "Pixel" is some specific program. Why are you capitalizing Pixel? Is it a software package? Because just a "pixel" from an image would not be capitalized. I assumed that somehow you'd gotten coordinates by scanning some image of a plot/graph and the coordinates were in pixels.

I still don't know what you mean by that third number.

Can you attach the original .spc files (zipped up). Why don't we just use those instead of messing with some text file with other, incorrect coordinates?

Answer 3

[David Goodmanson]

Now that the peaks in both pixels and cm^(-1) are provided, you can do a linear fit, which is very close.

p = [155.56834, 396.54987 624.69582,]
r = [280.05051  711.46459 1085.32579]
a = polyfit(p,r,1)
figure(1)
plot(p,r,'o-',p,polyval(a,p),'o-')
grid on
xlabel('pixels')
ylabel('inverse cm')  

      a = 1.7172   18.6607
i.e.
cm^(-1) = 1.7172*pixel + 18.66

so one cm^(-1) is worth 1.717 pixels.

It seems unikely in this case but suppose you knew for a fact that, because of the way the instrumentation is built and data is taken, cm^(-1) has to be strictly proportional to pixels, i.e. no offset like the 18.66 above. Then you could find the proportionality constant

c = r/p
p = [0 p];
r = [0 r];
figure(2)
plot(p,r,'o-',p,c*p,'o-')
grid on
xlabel('pixels')
ylabel('inverse cm')

     c = 1.7556

[caogqwerty123]

Could you attach the images of the graphs to see if your conversion method works? Please. Thanks a lot.

[David Goodmanson]

Hi ortizgd,

I guess I could, but your access to the files is much more convenient than mine. Defining

pixel file
p = pixels           (col 1)
Ip = intensity       (col 2)
inverse cm file    
icm = inv cm         (col 1)
Iicm = intensity     (col 2)

then with

figure(1)
plot(1.7172*p+18.66,Ip,icm,Iicm)
grid on

either the peaks willl match up or they won't.