How Google's book-scanner cleverly corrects for the curvature of an open book

One of the hard problems of bulk book-scanning is the distortion in the scanned images arising from the bowed center of the book as it lies open. Google's clever solution to this is to paint the book with infrared light, and then use two infrared cameras to generate a 3D model of the book, which can be used to correct the scans.

Turns out, Google created some seriously nifty infrared camera technology that detects the three-dimensional shape and angle of book pages when the book is placed in the scanner. This information is transmitted to the OCR software, which adjusts for the distortions and allows the OCR software to read text more accurately. No more broken bindings, no more inefficient glass plates. Google has finally figured out a way to digitize books en masse. For all those who've pondered "How'd They Do That?" you finally have an answer.
The Secret Of Google's Book Scanning Machine Revealed (via Memex 1.1)


  1. Thankfully they’re not using the “shred first, assemble later” technique that was featured at Verner Vinges’ Rainbows End…

  2. Alternatively, you could just open the book at a 90 degree angle instead of putting it on a flat surface. A while back there was a home-made book scanner on here that used that method.

  3. Why aren’t they using this? Seems like you’ll get a better result every time.

  4. Hm. Maybe that explains how they do their “dual layer” approach to PDFs, too– I’ve had many Google Books PDFs (and some Microsoft scanned ones, too) be in two layers– one with the text and all black lines barely visible, and one that is just text and black lines. When changing pages or opening the document, for just a moment you see the textless page before the text is overlayed on it.

  5. @2: Exactly what I was thinking … opening a book flat is basically asking to crack the spine.

  6. When the article title said that they “cleverly” corrected for page curvature, I assumed “cleverly” meant “wow! that was obvious now that you think of it.” Instead, I get “oh noes! science!”

    Very intriguing, i admit, but too much to swallow at 12 midnight.

  7. This is more interesting than actually useful.

    Any halfway decent OCR software would be able to accurately scan text with the little bit of curvature.

  8. Can’t they afford to sacrifice a book, get rid of the spline with a shear and use a much simpler scanner which would do a perfect job without all that fuss?

  9. Book scanners have existed for a *long* time, and have generally tended to perform some sort of geometric correction.

    The hard part is flipping the pages.

  10. #4: I’m a Bibliophile, but I think sacrificing one book for the sake of having it accessible to millions of people is acceptable.

    Unless that book is super rare or one of a kind of course.

  11. I’m wondering if this is for special case books? Existing high-speed book scanners have a vacuum-powered wedge that inserts between two pages, applies vacuum, then scans while retracting the scan head upward, always normal to the page, so text curvature isn’t an issue typically and they’re able to get up to 2400 pages/hour (vid reference: )

    maybe google’s method will offer faster scan rates if they somehow get the page flipping to be quicker than it is?

  12. Google should throw its $ and brains to use either CT xray or MRI in combination with a 3D OCR technique, that could be used to scan entire shelves, blocks of books or books rolling by on conveyor.

    CT xray and MRI are basically 3D OCR (organic ‘character’ recognition) tools.

    The biggest challenge will be developing the technique and calibration to resolve paper and ink and their variability. Also, I don’t believe current technology has fine enough resolution to distinguish pages.

    But conceptually, it seems reasonable.


  13. it should be noted that the google book OCR solution pre-dates the scan robot that others have posted a link to here, which is probably more accurate as well. I find the scan robot solution more elegant than googles, Though I wonder how fast it is compared to google’s scanner. The cost for the scan robot is around $107K.

  14. i can’t really say anything specific, because of scary nondisclosure agreements (i figure google is probably the entity to be afraid of if you’re going to be paranoid online). but i will say i worked as a book scanning grunt for google, and their method is much faster than any of these machines. this is also not a “new” method for them. just recently figured out by the public from these patents, i guess.

  15. @#10 – You’re describing Google’s alternate method for scanning books. They do that when publishers send in books to be scanned. Or so I’ve heard. Haven’t actually seen it being done myself.

  16. Why using glass plates that flattened each page is not very efficient? Compare to what Google use, I believe placing book in a V-shaped is much better for OCR because you have 100% curvature-free for the first place. You can also check out this V-Shaped book scanner at

  17. 1) Flipping pages instead of moving the entire scanner assembly uses less energy and is probably less prone to break down. When you deal in the volume that Google deals in, a 10% energy savings could be worth many millions of dollars.

    2) With Google’s method, it is click, flip, click, flip, click, flip all within seconds. The speed advantage must be great.

  18. Unfortunately Google’s book scanning can go wildly wrong. I would really like to know how this managed to pass muster and get posted in this horrific state:

    Hand-book of the Locomotive, Including the Construction, Running, and Management of Locomotive Engines and Boilers
    By Stephen Roper
    Edition: 14
    Published by E. Meeks, 1890
    324 pages

    Hand of archivist on pages:,M1,M1,M1

    Weird smeared text at bottom:,M1

    Pages in mid-flip during scan:,M1

    Sideways page:,M1

    Weird warped-edge page: (de-warp algorithm gone wrong?),M1

    Total scanning disaster:,M1,M1

    This archivist was either intoxicated or on drugs, just did not care about doing a good job. DO-OVER!

Comments are closed.