Can OCR (+artificial intelligence?) count how many of the same patterns?

Discussion in 'Embedded Systems and Microcontrollers' started by puppi, Dec 13, 2013.

  1. puppi

    Thread Starter New Member

    Dec 13, 2013
    4
    0
    Using Google Goggles that solves Sudoku as an example,
    Can a device like that (that seems to have both OCR and AI functions) count how many times each number shows up on a Sudoku game sheet, either before or after solving the game? (Solving is not necessary)

    There seems to be not much info on this out there, so any input would be appreciated! :) (Or recommendation for another forum, whatever...)
     
    Last edited: Dec 13, 2013
  2. tshuck

    Well-Known Member

    Oct 18, 2012
    3,531
    675
    Yes, yes, it can.
     
  3. ErnieM

    AAC Fanatic!

    Apr 24, 2011
    7,386
    1,605
    Solving a Sudoku game or counting up the occurrences of numbers are simple trivial exercises of procedural code.

    The extreme difficulty is developing the input software: OCR implies moving from the physical domain to the electronic computational domain: think physical paper in and PDF out.

    While I can't remember where I read his essay, Douglas Hofstadter devotes some time to the complexity of having a computing device recognize the letter A with the same ease as you or I. Just open MSWord and scroll down the list of fonts, then consider that a handwritten input is also valid.

    You or I can process such input "without thought."

    A computer needs "artificial" guilds and boundaries to help it in it's search. A #2 pencil, block letters only, insert this direction, some mechanical aid to guild it thru the real world to do a task you or I don't even have to think about.
     
  4. Potato Pudding

    Well-Known Member

    Jun 11, 2010
    684
    92
    But it is defineitely done.

    It is fascinating stuff. I talked to a grad student import from Poland who was doing Optical Recognition and counting for medical diagnostics. They put samples under a microscope to identify, characterize, and count tissue and phage cells.

    In some ways that is simpler than an alphabet and vocabulary.

    Alphabet is one problem. Things like license plate recognition are relatively trivial with the standardized font. Even capture and correcting for aspect from real world contexts is not too bad.

    Now you want to get tricky.

    IBM is looking at what it will take to read doctors handwriting! They want to solve the problems involved in scanning old medical charts and into a database that might be the research bonanza of the century.
     
  5. puppi

    Thread Starter New Member

    Dec 13, 2013
    4
    0
    Thanks! So, since the input technology is already made by Google (for Google Goggles, at least to input pre-printed numbers off Sudoku), do I just need to find a similar device that uses the technology and add a code to count numbers?
    Do you have any recommendation on such devices?
     
  6. Potato Pudding

    Well-Known Member

    Jun 11, 2010
    684
    92
    Any kind of image capture might work. Google Glass is an interesting image capture tool.

    There are steps involved.

    Image correction > OCR > Text > and you can count characters in text.

    You should be able to find most of those ready made if you do the research.
     
Loading...