How can I find a certain shape within an image?

I have a script that takes values from these velocity curves, but I'd like to automatically calculate the coordinates for each of the cyan crosses.

This is the mask I'm currently using. I've tried using normxcorr2 to match up a cross template with the crosses but it seems to get confused when the crosses are right next to the curves.

How should I go about this?

Answer 1

[DGM]

Is it necessary to use normxcorr? The content seems reasonably seperable by color.

inpict = imread('crosswav.jpg');

mask = inpict(:,:,1)<20 & inpict(:,:,2)>=210 & inpict(:,:,3)>=210;
S = regionprops(mask);
C = vertcat(S.Centroid);

imshow(inpict); hold on;
plot(C(:,1),C(:,2),'k+')

That said, I sure hope that the actual images aren't tiny jpegs. Don't let MATLAB save images as JPG if you ever intend to post-process them. It uses abnormally low-fidelity parameters for JPG encoding.

[Samuel Harvey]

Thanks! This is pretty much what I ended up doing and it works great.

Answer 2

[Image Analyst]

What is your actual starting data? Do you have a 1-D signal (y values and maybe x values too)? Or is all you have an image with colored boundaries burned into it? If all you have is the image, then how did it get created? Who created it? Is it possible to write out the colored curves you show on the image in a .mat, .csv, or .txt file?

If you have the y signals, I'd just find the peaks with findpeaks,

[peakValues, indexesOfPeaks] = findpeaks(ytop, 'MinPeakDistance', n); % n is approximately half the distance between major peaks.

then find the separation

ySeparation = yTop - yBottom;

then "fall down" the left side of the peaks until the separation begins to get larger

Like for one peak

for k = indexesOfPeaks(1)-1 : -1 : 1
    if ySeparation(k) > ySeparation(k+1)
        cyanIndex = k;
        break;
    end
end

Repeat for the other peaks. This will give you the location where the separation is narrowest just to the left of the big peaks.

[Samuel Harvey]

Unfortunately yes, I only have these images with the colored lines. We use other software to create them from Doppler velocity imaging but it doesn't have the functionality to export the results in a numeric format yet.

[Image Analyst]

Then try this (untested)

[r,g,b] = imsplit(rgbImage);
mask = (r == 0) & (g == 255) & (b == 255); % Find cyan blobs.
% Find centroids of cyan blobs.
props = regionprops(mask, 'Centroid');
% Get all (x,y) centroid coordinates into a nice 2-D matrix.
% xCentroid in column 1, yCentroid in column2.
xy = vertcat(props.Centroid)