Problem with col2im () function

Hello

I have a problem regarding the col2im function !

I have used the the im2col () with the property 'sliding' and a 5x5 kernel and so I have a 25 x 1000 matrix now !

I need to transform it back to an image but as I read the col2im() does not allows input to be a matrix (strangely) ! Can someone explain me why I have the following error !

 "To RESHAPE the number of elements must not change."

My input matrix (im2col) is exactly the same size as the output (col2im). I read about the advice not to use this function but know have implemented it in a large workflow and I would rather not change it !

I just need to reconstruct my image.

Thanks in advance

[Dimitris M]

Hello again

Can someone from Mathworks advice me on the subject ! This function seems to crash for no reason ! Even the documentation is not very helpful in this case !

Answer 1

[Image Analyst]

This function is rarely used - probably because it's so confusing and rarely needed even if someone did understand it. In my mind, they should be deprecated. If you're using a sliding linear filter, why not just use the much simpler imfilter() or conv2()?

[Dimitris M]

Hello

Thank you for the reply ! The thing is that I am using a kernel and the I am applying a non linear filtering process on the its values. So practically I need the values of the sliding window and do the following processes over them

1) sort them in ascending order

2) compute their absolute difference

I have done all the using the im2col function and processing the exported elements in a column wise manner but know I need to reconstruct them image. I am not really sure if it easy to construct an algorithm to do the same thing using imfilter() !

Is there another way to reconstruct my image using the data from im2col (with sliding property) ???

Thanks again

[Image Analyst]

I still don't see why you need that function. I wouldn't use it. In fact MATLAB has a rank filter like you're doing. So if you have a 3x3 window, you'd have 8 absolute differences with the center pixels. You can do this with 8 calls to conv2(). But what do you do then with the 8 absolute values of the 8 differences? What gets assigned to an output image?

[Dimitris M]

After some research in the internet I found this - with a small modification I was able to finally run my code !

http://charmie11.wordpress.com/2011/10/24/im2col-and-col2im/

[Image Analyst]

I still say it's an unnecessary and obsolete function. If you want to do non-linear filter, as you have said, then the function to use is nlfilter().