Thank you for your detailed explanation. I had basically two batches of logos, one scanned from a very old scanner and the other from a newer one. The driver of the old scanner did not have (or maybe I did not notice) the descreening function so I have to stand that if possible. But the newer one, I have tried to use a few descreening parameters but it doesn't seem useful. In fact, I do not even understand the meaning of the parameter of this function.

For resolution, I have tried 72dpi but seems it do not work. What resolution, in your experience, should be used?

I am interested in the theory of CYMK. I think if I know the theory more then I may be able to do something about it. Do you mean that each perceptual pixel is made of a 2x2 pixels composed of the four components? Is there any way to tell which 4 pixels should be grouped as one?


cate adams wrote:

I don't know if this will help, I am a photoshop user sharing this address, I've used gimp a bit and find it wonderful. it's amazing to get free software that seems to do everything photoshop does and more. i haven't scanned in gimp.

I do what you are trying to do regularly in photoshop (so maybe in gimp or other scanning software) when you scan you can choose to descreen the image as you scan. you choose between newspaper, magazine and art magazine etc. depending on how course the printed screen is.

You are getting this "dithering, anti-aliasing and "texture" of publication paper" effect because when the image is printed in a magazine the colour is made up of a fine dot screen of cyan, magenta, yellow and black, (cmyk) to make up all the colours you see, so even if they appear flat they generally aren't. for example fire engine red is 100%magenta and 100%yellow, green is percentages of cyan and yellow and so on.
Your scanner is picking up each little dot so you get a pixel of cyan and a pixel of yellow and it dithers it to try to get the inbetween colours.
I have found descreening the best way, i used to try it with filters like you are, getting pretty bad results as you are. If you can't do that you could try scanning on a much lower resolution so the scanner can't see the dots. It might sound funny but it works. You might sharpen the edges later if you need to though as you've discovered you have to be careful of getting a halo.
cheers cate

From: "Cheung Koon Tung, Kent" <[EMAIL PROTECTED]>
Organization: City University of Hong Kong
Date: Sat, 03 Feb 2001 11:56:00 +0800
Subject: Re: Scanned Image Processing
Hi all,

Thank you for all the nice guys who have given me input. Here are my experiences about your suggestions:

> Perhaps doing Posterization (Image/Color/posterize)
1. It is hard to determine the paramter of posterization. Setting it too high is not useful. Unfortunately, setting is low will cause some pixels to vanish. Besides, some noise pixels with very different colours from the perceived one will be produced.
> Check out Image Magick, it has some nifty color management
> tools.  Look at the things you're scanning with a magnifying glass and you
> should see many different colors and a halftone screen dot
> pattern.
2. ImageMagick is a very good program that allows infinite programability that provides more or less the same as GIMP or other image processing applications. If I had found the way to process my images in general, I will be able to write a script for GIMP or ImageMagick. Unfortnately, I am still searching for the method.
> I haven't tried this, but maybe fiddling with indexed mode and
> limiting the number of colors would help.
3. If I index the colors with the standard palette, then some of the original colors will be lost. However, creating palettes for all images I am going to process is a nightmare.

> You might want to try the non-linear filter (nlfilt).
4. I think it doesn't help much. I don't even notice the changes to the image! Of course, I have changed the parameters but not any one set of parameters are satsifactory.

5. Is there any standard routine to remove the three artifacts I mentioned, namely, dithering, anti-aliasing and "texture" of publication paper? Since these are produced by standard routines, I expect that there might be standard methods to undo the effects.

6. I have another thought: is it possible to let the user to specify a small number of colors that the image should have. Then, program GIMP to change the pixels which don't have these colors to its correct colour. The interface is certainly possible using Gimp-Perl. However, the way to determine the correct color is difficult to determine. Any idea?

Thank you for all your kind help.


Reply via email to