Reducing Matte Artifacts in Scanned Photos

  • 1
  • Question
  • Updated 2 years ago
I use the following filters in the order listed to improve my scans of Matte Images:

Noise Dust & Scratches: Radius: 3, Pixels, Threshold: 12 Pixels
Reduce noise: Strength: 8, Preserve Details: 60%,
Reduce Color Noise: 45%
Median Filter Radius: 3 Pixels
Median Filter Radius: 3 Pixels
Gaussian Blur Radius: 2 Pixels

Fortunately, most of the photos were taken with a low quality camera, and I scan them at 1200 Pixels/in. Consequently, while these settings blur the image slightly, the overall image quality is much better.

In the attached image, the right end is 'as scanned' from the 'white' edge of a photograph. The two grayscale images were created by using the :
Enhance-->Auto levels.
The filter sequence above was applied to generate the center panel.
These were then converted to black and white to exaggerate and emphasize the Matte matrix. In

I fully believe that due to the periodic nature of the matte finishes, a matte removal filter could be designed. Note: This is only one of several matte patterns that I have seen.

Photo of John Priest

John Priest

  • 4 Posts
  • 0 Reply Likes

Posted 2 years ago

  • 1
Photo of Chris Cox

Chris Cox

  • 20280 Posts
  • 766 Reply Likes
There is no perfect way to clean up the matte texture.
An FFT filter would be closest, but you would still limit the cleanup to the matte area and not apply it to the entire image.
Photo of John Priest

John Priest

  • 4 Posts
  • 0 Reply Likes
I agree, especially since the texture can be hidden by the photograph itself and by photo damage. My thought on fairly simple filter would be to use a 2-D Fourier transform filter to extract the matte pattern. Even better would be to follow up with a deconvolution to improve the image using the extracted mask.

If we could get images of all of the possible matte patterns, the problem might be simplified. In the photos I have examined the two main patterns appear to equilateral parallelograms or equilateral hexagons, both of which will 'close' pack.