Enhancing an RGB image to identify a feature

I have an image which has a faint line that is NOT continuous. I want to find the pixel intensities along the line so that if there is a drop in intensity i can identify it as a break in the line. For example, in the green apparently "uniform" line below, there are supposed to breaks in between. I need to find the pixel intensities along that line to find the break. I tried using the image tool to see the pixel intensities, however, the breaks are too faint to be noticed.

I tried enhancing the image by using imadjust (both in rgb and converting to grayscale then trying imadjust) but it didnt help. Any other methods to enhance the line so that the breaks are visible, or atleast the pixel intensity variations along the line are significant?

Any help is appreciated, thank you!

[Walter Roberson]

You might be able to work something along these lines

filename = 'https://www.mathworks.com/matlabcentral/answers/uploaded_files/843265/image.jpeg';
img = imread(filename);
imshow(img)

imshow(img(:,:,1)); title('r');

imshow(img(:,:,2)); title('g');

imshow(img(:,:,3)); title('b');

lab = rgb2lab(img);
min(lab(:)),max(lab(:))

ans = -20.3143

ans = 91.8858

whos

  Name            Size                 Bytes  Class     Attributes

  ans             1x1                      8  double              
  filename        1x80                   160  char                
  img           274x767x3             630474  uint8               
  lab           274x767x3            5043792  double              

imshow(lab(:,:,1), []); title('L');

imshow(lab(:,:,2), []); title('a');

imshow(lab(:,:,3), []); title('b');

Notice that 'a' appears to distinguish the line we want in a way that hints we might be able to binarize. But at what values?

imagesc(lab(:,:,2)); colormap(parula); colorbar(); title('a colorizeed');

abin = lab(:,:,2) > 0 & lab(:,:,2) < 14;
imshow(abin); title('a bin')

If you change to < 15 instead of < 14 then a lot more shows up in white, and you can see the line doesn't show up long enough, so somehow the hints from the imshow(, []) do not carry over well to the imagesc(); colormap('parula'). Perhaps using the data cursor on the imshow(lab(:,:,2),[]) would help find a range of values that is more usable.

Answer 1

[yanqi liu]

clc; clear all;
close all;
img = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/843265/image.jpeg');
J = rgb2ycbcr(img);
im = mat2gray(J(:,:,1));
im = imadjust(im, [0.45 0.8], [0 1]);
ed = imbinarize(im,'adaptive','ForegroundPolarity','dark','Sensitivity',0.65);
ed2 = imclose(imopen(ed, strel('line', 11, 0)), strel('line',19,0));
ed2 = bwareafilt(imclearborder(ed2), 1);
[r, c] = find(ed2);
p = polyfit(c,r,3);
x = linspace(1,size(im,2),2e2);
y = polyval(p, x);
figure;
subplot(2, 2, 1); imshow(img);
subplot(2, 2, 2); imshow(im, []);
subplot(2, 2, 3); imshow(ed, []);
subplot(2, 2, 4); imshow(ed2, []);
hold on; plot(x,y,'r-', 'LineWidth', 2);

% find the pixel intensities along the line
g = rgb2gray(img);
g2 = zeros(size(g));
for i = 1 : length(x)
    ri = round(y(i)); ci = round(x(i));
    ri = min(size(g,1), max(1,ri));
    ci = min(size(g,2), max(1,ci));
    g2(ri,ci) = 1;
end
gs = g(logical(g2));
figure; plot(gs);

Answer 2

[Image Analyst]

Did you try calling improfile(). Have the user draw a line, extract the colors along the line in each color channel, and then plot them.

[Manas Pratap]

It helped to an extent. Is there a way to freedraw to create a profile? rather than just a straight line?

[Image Analyst]

Yes. See and adapt my freehand drawing demo.

Also, is the faint line the mostly vertical one, or the mostly horizontal one, or do you want both? Is the panel always going to be situated at that angle, like it's held in a jig or something? Or is it some lousy handheld cell phone snapshot?

Can you do anything to improve your image capture situation? Usually making a simple change on the image capture situation will make lots of complicated algorithm development for analysis unnecessary.

Have you tried looking at the thing in non-visible light, like infrared or ultraviolet?

Have you tried stdfilt() or imtophat()? How about anisotropic diffusion?

Basically your image is virtually useless because you have horrible JPEG artifacts. Never use JPG when you're planning on doing image analysis. Can you supply a non-corrupted image in PNG format?

[Manas Pratap]

The mostly horizontal line. It is a handheld phone snapshot,but the paper on which the line is on is suspended in a gel matrix, so adjusting the angle might not be viable.

The idea is to use something in the visible spectrum itself, perhaps changing the background light might help.

I havent tried the functions you've mentioned, but I will give them a go.

unfortunately I cannot do anything to improve the image provided to me, as it was just given to me.

Thanks for your help!

[Image Analyst]

Then if your image capture person can't (or is unwilling to) do anything to help you solve their problem, I suggest you give them a human-assisted solution like I originally suggested. Have them hand draw the line and then you can do whatever you want with those coordinates.