hello, am working on my project which involves detection of cracks on rail using image processing algorithms on matlab. please i need some help on the matlab code that will enable me extract crack on rail. the rail images am working with are attached below. thanks you so much
Start with an "edge detector". For this problem, you would only need to apply this in the horizontal direction (assuming the cracks always run vertically), which simplifies things greatly. So no need to apply 2D convolution kernels here, you could get by with 1D filters, for first-pass processing at least.
The edge detector is basically a high-pass filter. A simple difference is a limiting case of a high-pass filter. Use matlab's Y(m,:) = abs(filter(b,a,X(m,:))) command, where X(m,:) is the mth row of your image, Y(m,:) is the mth row of your horizontal difference image; a and b are the coefficients of your filter, start with the difference filter with a = 1.0 and b = [1,-1]. (Try a higher order high-pass filter if that doesn't work.) Finally, take the absolute value of the difference filter output (because you don't really care about the sign).
You now have a difference image.Next, apply a threshold (lim) to create a binary image. e.g. use
Y(find(Y>lim)) = 1; Y(find(Yp_lim in any given column.
Anyway, give that a try.
I think image feature extraction algorithms such as SIFT might be applicable. I suggest you apply SIFT to the image with Crack and the image without crack then compare the the points detected. I am not sure if SIFT is available in Matlab, but SURF is a good replacement and it is available in matlab.
Start with an "edge detector". For this problem, you would only need to apply this in the horizontal direction (assuming the cracks always run vertically), which simplifies things greatly. So no need to apply 2D convolution kernels here, you could get by with 1D filters, for first-pass processing at least.
The edge detector is basically a high-pass filter. A simple difference is a limiting case of a high-pass filter. Use matlab's Y(m,:) = abs(filter(b,a,X(m,:))) command, where X(m,:) is the mth row of your image, Y(m,:) is the mth row of your horizontal difference image; a and b are the coefficients of your filter, start with the difference filter with a = 1.0 and b = [1,-1]. (Try a higher order high-pass filter if that doesn't work.) Finally, take the absolute value of the difference filter output (because you don't really care about the sign).
You now have a difference image.Next, apply a threshold (lim) to create a binary image. e.g. use
Y(find(Y>lim)) = 1; Y(find(Yp_lim in any given column.
Anyway, give that a try.
Hello. Hugh Kennedy's answer is a good start. As you can see, it all depends what your model for a crack is in the image. If a crack is a straight line, then you should try to detect straight lines in the image as candidate cracks. There are some rotationally invariant edge detection methods around, which might be useful for you. They are not hard to code in matlab. Otherwise, you could try Hough transform.
Hi Ismail i am also doing my research on automatic crack detection... i need your help ..
- Badges
- Science topic