Hello,
I have some scanned images. Such images come from a booklet that has been impregnated with oil or sometihng similar (accidentally). For this reason, while looking at the original pages (and at the scanned ones) you can see both the page you’re looking at and the page on the opposite side of the sheet. Some images are present in this booklet and I would like to remove the text that is on top of those (that is coming from the oher face of the sheet and is visible beacause of the oil that impregnated the page). I could correct each scanned image “pixel by pixel” manually, but it would take a huge amount of time. Do you have any suggestions on how I could solve this problem maybe usin GIMP? Thank you! In the sample I attach, you can see a part of one image and some test that is coming from the other side of the page: I would ike to eliminate that text.
Before you spend too much time with Gimp or equivalent: if you scan with something dark behind the paper (instead of the usual white-covered press you find in most scanners), there is no light reflected back through the back side of the sheet, and this makes the print on that side a lot less visible. The more difference, the easier it is to increase it with Gimp…
Thank you very much for your reply. Following your suggestion, I did some tests scanning both color and greyscale:
I see that using a black background the text on the other side is less evident (as a side effect it seems I loose some detail of the image, but I’m not sure). Is it possible to further improve the image?
My first scanner many years ago was made by Agfa and had a black cover exactly to prevent the print on the other side to become visible. When I scan documents printed on thin paper I still use a black sheet with my new scanner.
In case of @egev probably the paper has become very transparent and the print colour might have diffused through the paper sheet.
With the Purple background and the Mono Mixer, it is possible to almost extinguish the text from the back:
Which make me wonder if a lighter but strongly colored background (green screen) wouldn’t allow more separation…
also try adjusting the levels when you scan. tweaking the levels you might be able to remove the background text.
There is another option that would prolly be an exercise in frustration in trying to get the images to align, but making a mask of the next page, flipping it and allying it to the first page might work to some degree.
After that this would be ideal for machine learning to tackle.
Thank you all.
@AdmFubar: At first I tried to scan both sides of the page, then aligned those and subtracted one from the other, but I was not able to achieve a satisfactory result, however I would like to try again, but I must think about a good way to get the two scanned images aligned in the best way.
@Ofnuts: For sure using a dark background is helping me. Here is the image scanned with a green background:
Just for info, this is the green I used (as seen by the scanner):
However, the best result I obtained using Mono Mixer is the one with the purple background. Using the green one I was not able to obtain a better result. Here is the result after using Mono MIxer (and purple background):
For me (I am not an expert for sure) now the problem is to achieve a good contrast between the drawing and the background. Ideally I would like to achieve a white background and black drawing but I’m not sure it can be achieved. How would you proceed?
Did some color sampling on areas o fthe “purple” scan that correspond to the drawing on the front, characters from the back, and areas where they overlap:
- The overlap areas have a rather obvious smaller difference between front red and green channels
- The front drawing has less difference between red and blue channels than the back characters (so it"s more purple…)
However these differences are tiny and I don’t see a process in Gimp to really take advantage of this. Extracting this kind of data would require some code.
I wonder if a reproduction using really flat light instead of the straight on light of the scanner would bring out the front better.