How to proceed to identify shadows in an image?

Question

1 vote

Hi. I want to detect shadows . I tried two algorithms which did not work for all images.

1) finding (R+G+B)/3 for a single pixel and computing mean for the image by adding all pixels and divide by no of pixels. Those values below mean are considered as shadows

2) finding new image x using

x(i,j)=(4/pi) .* atan( (double(I(i,j,1)) - sqrt(double(I(i,j,1)^2 + I(i,j,2)^2 + I(i,j,3)^2))) / (double(I(i,j,1)) + sqrt(double(I(i,j,1)^2 + I(i,j,2)^2 + I(i,j,3)^2))) );

then apply otsu's threshold to detect shadows.

Is there any other method to detect shadows

1 Comment
Show -1 older comments Hide -1 older comments

Walter Roberson on 11 Feb 2012

I added spaces to the formula to try to make it more readable.

Sign in to comment.

Sign in to answer this question.

Follow Question

Answer 1

Image Analyst on 11 Feb 2012

2 votes

Your proposed method 1 does not find shadows. All it tries to do is find the half of the pixels in the image that are below the mean. Only in certain rare and special images will exactly the darkest half of the image be a shadow. And I'm not sure what the point of taking the arctangent was in method #2 where you're taking the arctangent of the red channel minus the luminance divided by the red channel plus the luminance.

Try this: <http://groups.google.com/group/comp.soft-sys.matlab/search?hl=en&q=shadow&start=10&sa=N&hl=en&>

And also search for shadow on this site: http://www.visionbib.com/bibliography/motion-i762.html#Background%20Detection,%20Background%20Model There are lots of papers on it there.