Daniel Reetz, the founder of the DIY Book Scanner community, has recently started making videos of prototyping and shop tips. If you are tinkering with a book scanner (or any other project) in your home shop, these tips will come in handy. https://www.youtube.com/channel/UCn0gq8 ... g_8K1nfInQ

Will increased megapixels improve OCR?

Convert page images into searchable text. Talk about software, techniques, and new developments here.
Post Reply
Marcus Eastty
Posts: 9
Joined: 13 Apr 2015, 01:45
Number of books owned: 0
Country: United States

Will increased megapixels improve OCR?

Post by Marcus Eastty » 24 Jun 2015, 03:24

I guess my question is more of a, "how does the software recognize the characters?"

I kind of assumed it looked for blotches of contrast-colored pixels that resemble characters of a given language.... But is it looking at the whole image and recognizing characters, i.e. "reading" the page like a human would, or is the analysis of the pixels directly from the bytes of data stored in the photo?

I ask because the camera I use is a camera phone, and the app I use recently released a version where the program takes 10 photos from a 13 MP sensor in succession and stitches the "best" 4 of them to create a "50 MP" photo. The metadata from these photos has the corresponding resolution. I can maybe see a difference in the photos I've seen (though I don't think I should be able to tell because my laptop's screen resolution is less than what the images can provide).

Anyway, would OCR improve in accuracy with this method? DPI of input into scantailor would increase, and so I would expect the output to also increase. From there, I wouldn't be able to see a difference, but would ABBYY be any better able to recognize characters?

Another shooting option would be a "clear image" option, which is another 10-shot stitching for "improved clarity," but with the same 13MP resolution. Any ideas?


Posts: 243
Joined: 19 Mar 2013, 14:55
Number of books owned: 0
Country: UK

Re: Will increased megapixels improve OCR?

Post by cday » 24 Jun 2015, 04:07

Abbyy gives advice on image resolution and colour mode in its manuals and online, if you Google something like "abbyy finereader ocr dpi".

The following text from this link is typical:
What image resolution is the best one?

For regular texts (font size 8-10 points) it is recommended to use 300 dpi resolution for OCR. The reason is that all ABBYY technologies are tuned for that resolution.

If scans have a smaller resolution, for example 200 dpi, then 10 point font will be too small. To compensate the “missing” pixels, the image will be scaled internally (up to 400 dpi). Low image quality (i.e. resolution) may lead to not only quality but also speed degradation as uncertainty in character picture produces more recognition variants to process.

For smaller font text sizes (8 points or smaller) we recommend to use A 400-600 dpi resolution.

Colour, grey or black & white images - What is recommended?

ABBYY Technologies use colour information for detecting areas and objects on the image.
So, if complex layouts have to be processed, it is recommend to use colour or at least, grey scale images.
The character recognition is always executed on an bi-tonal image, that only contains black & white. But to archive a good OCR result it is important to generate a good, suitable binary image. ABBYY technology is not just using “simple” binarization, but “adaptive binarization.”
So you might see some improvement in accuracy from enhanced images but some comparative tests would be the best answer.


The above link is worth reading in full, and at the end also links to other relevant content.

Posts: 383
Joined: 01 Jun 2014, 17:04
Number of books owned: 1000
Country: United States of America

Re: Will increased megapixels improve OCR?

Post by duerig » 24 Jun 2015, 14:56

As cday says, the rule of thumb is that you are aiming for 300dpi. There is some benefit to going a bit higher (400dpi), but there is almost zero benefit to going higher than that. So if I were you, I would stick with the super-resolution your camera gives you knowing that it will almost always be good enough. This gives you the luxury of not having to ever worry about zooming in to maximize your DPI.

But now I really want to get my hands on one of these cameraphones. In another thread, boredland has been working on getting Spreads to work on Android devices. With one of these cameras, you could have the controller, the camera, and everything all in one package. That would be just awesome.


Post Reply