Descreening an Image

There may be times when you are required to remove the halftone “rosette” pattern from images that have been scanned from printed media. This is done to prevent the previous dot pattern from interfering with the new pattern that will be generated when the new separations are made. Another important reason to do this is to reduce file size and to restore a photographic look to the image.

When performing this process for printing, it’s assumed that you have permission to reproduce the image from its copyright holder. It may simply be a case where the original is lost or extremely difficult to obtain. The same conditions apply if you were to use the image as part of a photo composition, although copyright laws do allow images to be used without permission as long as they don’t exceed a certain percentage of the entire image.

That said, this tutorial explains how to convert the rosette pattern into a random pattern that resembles film grain with a minimal loss of detail.

Many are under the false impression that the way to do this is through blurring. Unless the image is blurred to the degree that fine detail is lost, it simply softens the existing pattern. What we need to do is to “break up” the pattern so that it’s no longer a regularly occuring texture. And the filter that excels at this is Diffuse.

Diffuse works by randomly moving pixels within each “sample”, which is exactly what we need to do to rearrange the rosette pattern to look irregular, similar to film grain. The trick is to find the optimum resolution at which to apply the filter to break up the pattern while preserving the detail.

We then apply Add Noise to enhance the effect of the Diffuse filter. Usually, 3% of uniform noise works well. Applying it as “Monochromatic” helps to keep down color noise.

To take the edge of the resulting “sharp” pixel texture, we then apply Gaussian Blur at a very low percentage, about 0.5%.

All of this results in texture similar to film grain. And since film grain is often perceived as noise, we can treat it as such and reduce it with Median, so this is the final filter we will apply. A radius of one pixel usually does the trick, as we don’t want to obscure too much detail.

Step by Step:

1. Begin with a pre-screened image (this is a crop of the original image at its orginal resolution.)

2. Assuming it was scanned at 100%, and that it has the typical line screen frequency of 133 lpi, resample the image to about 150% of its original size using Image > Image Size… (Command + Option + I). Some experimentation may be necessary to find the optimal resolution, but applying this effect to the original image often results in a substantial loss of detail.

3. This effect is best applied as a “Smart Filter” so that it’s possible to experiment with different filter values to obtain the optimum effect. To do this, convert the current layer into a “Smart Object”:

4. Apply Filter > Stylize > Diffuse…

5. Apply Filter > Noise > Add Noise…

6. Apply Filter > Noise > Median…

7. Finally, downsample the image to its original resolution (66.66… %), or the size and resolution of your choice. Your Layers panel should now look like this…

And the resulting image like this…

If the image had been simply blurred enough to obscure the rosette pattern, it would have looked like this:

Moving Forward, you can now apply noise reduction to smooth out the grain if desired. Another trick is to increase the Median filter slightly (2 to 3 pixels is usually a good value) and apply Unsharp Mask to sharpen the image. The result may not be to your liking, but it is an option if the grain is distracting.