I have a number of jpg images whose quality I would like to improve. It is planned to both print them and use them in digital format. I guess to use the GIMP or another programm that I can install in my Linux. I expect you to advise me a tutorial with relevant information. I have reduced original images with ImageMagick's convert
command for posting here. These are scans of award documents from the Second World War in Russian.
Answer
The simple technique
Use the Curves tool.
If you look at the image histogram:
- The hump (1) is a peak frequency of dark pixels and is your text
- The hump (2) is a peak frequency of light pixels and is your background
- No pixels beyond (3) because your background is rather dark.
So with Curves, you make a diagonal between (A) (everything darker will be pure black) and (B) everything lighter will be white. You can make the curve follow a slight "S" shape to improve contrast.
An inconvenient of that method is that it is limited by 1) color shifts and 2) darker parts of the background that make it hard to handle the text correctly.
A more efficient technique (Gimp 2.10)
- Decompose the image into frequency components: Filters>Enhance>Wavelet decompose. YOU get alayer group called "Decomposition". Each layer of this group contains a part of the image. The top layers are the fine detail (the text) while the bottom layers are the general image (the paper).
- You can easily clean most of your background and make it white by bucket-filling the "Residual" layer with white. You can also tryo to bucket-fill the layer immediately above it.
- You can add "sharpness" to you image by duplicating the "Scale 1" layer (and sometimes the "Scale 2" one as well).
- However, your text is not very dark, so create a new layer from all this (Layer>New from visible) and move it to the top of the layer stack.
- Then you can use Curves as above, making a diagonal where the bottom point is where the histogram stops to the left (you can leave the other point in the top right corner). You can also add some contrast with an S-shaped curve.
No comments:
Post a Comment