MATLAB: detecting non-continuous vertical streaks in greyscale image

Question

In the attached image there are periodic vertical streaks which are random along the x axis and vary along the y axis in their intensity.
Any suggestions on how to detect these?
Ideally I would like to detect these with an output of a binary image showing the streaks (I imagine it will end up looking like a bar code).

Thanks!

Answer 1

Photon's answer is very good: he suggested alongated vertical edge filters to capture the vertical edges of the streaks.
However, if you are interested in locating the streaks themselves and not their edges, you might consider a slightly different approach:

I propose to, first, eliminate the "DC" component, so that the streaks will "pop out" of a roughly constant background, then use a vertical sum to locate them and produce a mask.

Here's a sketch of the code:

img = im2double( imread('http://i.stack.imgur.com/SqZrf.jpg') ); %// read the image

Use a horizontal filter to get an estimate of the local "DC", that is, an image where the streaks are gone:

dc_est = imfilter(img, ones(1,31)/31, 'symmetric' ); 

Looking at the difference between the image and the estimated "DC" should make the streaks pop out and easy to threshold using a global threshold:

global_thr = 0.025;
mask = ones(size(img,1),1)*(mean(img-dc_est,1)>global_thr);
figure; imshow(mask);

And heres' the result (you might want to change the threshold value and see how it affects the result):

The estimated "DC", dc_est looks like:

---

If you are after a more elaborate adventure, I recommend that you explore this work: I. Horev, B. Nadler, E. Arias-Castro, M. Galun, R.Basri Detection of long edges on a computational budget: a sub-linear approach (SIAM 2015). This method is aimed at finding these elusive edges and ridges in noisy intensity images.

Answer 2

You can use a simple horizontal filter to collect information about these lines:

I=imread('lines.jpg');
I=double(I)*(1.0/255.0);
J=filter2([-1 0 1;-1 0 1;-1 0 1],I);
s=sum(J);

Which results in the following 1D signal for the image.

You can then find the right threshold for you on this signal, and use the index as the line position.