Intel® Integrated Performance Primitives
Deliberate problems developing high-performance vision, signal, security, and storage applications.

Filtering Noise From An Image

Jordan__Kyle
Beginner
773 Views
I am trying to make an application that removes the noise from an image. And I can't figure which function in the IPP would serve best for this. It has to be as precise and as good as possible. I haven't had the time to check any of them out ( there are quiet a bit ), but does anyone have a suggestion? I know the IPP has what I need ( the IPP is quiet amazing ), and like I said, I've looked through the functions there are for filtering, but I haven't had time to tinker with them to see which is best.


The reason I have to make it as precise as possible, is because I am working with old documents that have been handled A LOT. I need to be able to erase the columns and rows that are there, but make sure I don't delete any of the print. All of the scans of documents are the exact same size and the lines are in the same place. So I got one document, copied the image twice on my computer, removed the lines from one document and subtracted it from the first image. That way everything else was erased except for the rows and columns. Now I just need to do some filtering to get the rows/columns clean so I can feed the image to a "machine", if you will, and it can delete the rows/lines specified from the image I filtered on the rest of the documents I send through it.


If anyone has any suggestions for the filtering function, that would be excellent and a HUGE assistance to me!
0 Kudos
5 Replies
Jordan__Kyle
Beginner
773 Views
Surely someone here can suggest something?
0 Kudos
Vladimir_Dudnik
Employee
773 Views

Hello, image de-noising is a big topic, I'm not sure one can suggest you something as simple as just call of one IPP function.

I may suggest you to try ippiFilterWiener which is usually may be applied in context of image denoising.

Regards,
Vladimir

0 Kudos
Intel_C_Intel
Employee
773 Views

The possible way is the deconvolution. IPP contains deconvolutiob functions for FFT and Lucy-Richardson algorithms. They eliminate the convolutional noise but you should try and choose the proper arguments: the kernel, number of iterations etc.

0 Kudos
levicki
Valued Contributor I
773 Views

If the documents are black and white (scanned as grayscale), and you need to keep only text, you may try to use threshold. If you have moire effect caused by scanning, that has to be removed manually by doing an FFT, then "deleting peaks" (bright spots located off-center in the frequency domain) and perform IFFT. If you have noise, the best method is to use Wavelet transform.

0 Kudos
kyzclockwise
Beginner
773 Views
Where can i find sample code on how to use these dconvolution functions?
0 Kudos
Reply