Enhancing text in scanned images

[crackhammer]

So I am in the process of scanning a book which is a mix of lot of images and text. I scanned the book using Epson Scan program and in home mode with text/art option @ 300 dpi tiff.

Now that the book is processed through ScanTailor, the text in the images doesn't look sharp. Is there any way that I can sharpen the images to enhance the text?

Thanks in advance.

[mr ploppy]

It'll be the de-screening option that makes the text fuzzy, but you'll need it for the pictures or you'll get a moire pattern. Best thing to do depends on what you have in mind for a final product. If it's OCR, scan the text separate without any de-screening. If you just want a set of pictures with text on them, scan them at double the size you want them to be without the de-screen, apply a gaussian blur to the images only, then resize the whole thing down 50%. That's pretty much what the de-screener on cheap scanners do in memory, but if you do it yourself you can undo any results you don't like.

[osnova]

While I responded in another thread, I can also put a couple of words here. To have the sharpest text in scans, scan into 300 dpi grayscale tiff with all processing turned off in the scanning app, process through ScanKromsator and upscale to 600 dpi B/W (SK is a pain to learn though). This is the only way to get razor sharp text that I've found.

[crackhammer]

@ mr ploppy,
Final product on my mind is, a pdf file with searchable text. The reason behind it is, I usually highlight the text of my interest and thats about it. I am not interested in perfect OCR.

@ osnova,
The program you mentioned seems to have loads of setting. I am pretty comfortable with SK, although I admit that I don't 'understand' all of its options. I will give a try to your program on my next project. Thanks for heads up.
The current book that I am scanning has loads of images (and they are necessary for the text in the book) So no black and white option for this book, has to be color.

[frostschutz]

Upscaling doesn't sound right to me...

If the text is blurry / small w/o postprocessing - scan with higher DPI. Otherwise just find the right threshold, i.e. turn it into pure b&w without gray.

[osnova]

Originally Posted by crackhammer

The current book that I am scanning has loads of images (and they are necessary for the text in the book) So no black and white option for this book, has to be color.

The advanced features of SK can break the page into zones and process these zones separately. So, say you scale the text up to 600 dpi B/W while leaving the pictures in 300 dpi color and then save to pdf from SK directly. I've seen this done by gurus of SK but have never done this myself

[osnova]

Originally Posted by frostschutz

Upscaling doesn't sound right to me...

Maybe I am using the wrong terminology but I have seen with my own eyes when so-so grayscale text turns into crisp sharp 600 dpi B/W text after SK. I actually used to do it myself quite often.

Under its hood, SK uses some advanced math algorithms (one drawback of this program is that it takes a long time even on the advanced hardware to work, I am talking hours at times). If I am guessing correctly, the author is some math guy from Russia.

[mr ploppy]

Originally Posted by crackhammer

@ mr ploppy,
Final product on my mind is, a pdf file with searchable text. The reason behind it is, I usually highlight the text of my interest and thats about it. I am not interested in perfect OCR.

For searchable text you will need to OCR it, so scan the images separately and put it together in a layout program to convert it to PDF.

[mr ploppy]

Originally Posted by osnova

Maybe I am using the wrong terminology but I have seen with my own eyes when so-so grayscale text turns into crisp sharp 600 dpi B/W text after SK. I actually used to do it myself quite often.

You might mean bitmapping, where it changes it to pure black and white (2 colours)? That would definitely need an increase in resolution for it to work, but for normal greyscale increasing the resolution would only increase the blockiness of any text.

[Penforhire]

Sounds like you should be able to find some scan/post-scan settings to improve it. Otherwise it is cumbersome but possible to improve such TIFF images in a good photo editor like Photoshop. There are filter tools like Unsharp Mask, image 'adjustments' like Threshold, and direct tonal adjustment (e.g. Levels dialog box).

[osnova]

Originally Posted by mr ploppy

You might mean bitmapping, where it changes it to pure black and white (2 colours)? That would definitely need an increase in resolution for it to work, but for normal greyscale increasing the resolution would only increase the blockiness of any text.

Yes, SK has selection for "binarization" algorithm and thresshold but there is more to it. The way I understand it, SK uses data in grayscale to extrapolate pixels for binarization and increased dpi. So, it is not just some dumb grayscale to B/W. When the conversion (300 dpi grayscale to 600 dpi B/W) is done, the "blockiness" of the text is reduced, not increased while blurry text becomes sharp.

It seems like magic, when the text that is in the shadow (say close to the spine), appears clear without a shadow and in increased resolution.

===

I also saw very good results from people processing scans through Corel PaintShop Pro but it is beyond my skills/abilities. They practically separated the text layer from the background layer.

[frostschutz]

Originally Posted by osnova

When the conversion (300 dpi grayscale to 600 dpi B/W) is done, the "blockiness" of the text is reduced, not increased while blurry text becomes sharp.

Sounds a bit like what this scaler does:

http://www.hiend3d.com/hq4x.html

While those examples look impressive, you should be aware that it's just eyewash. Such a process does not add real detail, and there are cases when such a filter actually makes things worse.

If you want 600 dpi, you have to scan 600 dpi. There is no other way.

[rkomar]

I had this same problem when scanning some of my old physics books. The problem is that the scanned text is usually dark grey rather than black, and the poorer contrast of E-Ink displays makes it hard to read. I ended up writing a Java application for selecting rectangular regions in the scans to leave as grayscale, and changing everything else to black or white. The SK program that @osnova describes sounds better than my home-built job. My only point is that you should convert the text to black and white to improve the contrast.

[frostschutz]

Originally Posted by rkomar

I ended up writing a Java application for selecting rectangular regions in the scans to leave as grayscale, and changing everything else to black or white.

I did that for a text that had mixed text/images once. Except I didn't write a Java application - I just used Gimp. Lots of manual labor but it's worth the effort when you can adjust the contrast of each image individually. And having everything but the images in pure black&white helps a lot getting the PDF down to managable size.

[rkomar]

Originally Posted by frostschutz

I did that for a text that had mixed text/images once. Except I didn't write a Java application - I just used Gimp. Lots of manual labor but it's worth the effort when you can adjust the contrast of each image individually. And having everything but the images in pure black&white helps a lot getting the PDF down to managable size.

I left the scans as grayscale data (with the text areas forced to be black (0) or white (255)). So, there was some savings in size because the black and white areas compressed better than the grayscale, but it wasn't huge. Were you able to store the text as monochrome (i.e. 1-bit) data for better size savings? If so, how? Could you save them as different layers on the page?