14

I have a database of Chinese character stroke animations. Each entry is basically a set of points which shows how and in which order a character should be drawn.

To make lookup easier, I would like to allow the user to manually draw a character with the mouse, then my software would propose some possible matches (like it's done on the iPhone).

I searched on Google but didn't find anything relevant. I'd like to know what algorithms or research paper I need to look at. I guess it has something to do with shape recognition, except that the order in which the shape (Chinese character) is drawn is also relevant. Any suggestion?

Chris
  • 368
  • 3
  • 13
laurent
  • 249
  • 2
  • 5
  • 2
    To the close votes: any suggestion on how the question could be improved? I realize it's a bit vague but this is because I don't know where to look (I don't know what's the name of the algorithm I'm looking for, although I think it probably exists). –  Apr 29 '12 at 13:04
  • 1
    If you haven't already Google for 'handwriting recognition' and 'gesture recognition' too, both of these are concerned (in part) with the process of drawing a character as well as the character produced. –  Apr 29 '12 at 16:55
  • If the number of strokes is always the same, you could group the characters in the database according to #strokes and directions (let's say 8 directions, and then a group would be represented by a sequence of numbers in the range 1..8), and then run some sort of shape matching on the narrowed-down group. –  May 17 '12 at 05:18

2 Answers2

5

You were looking for scholarly articles, I am surprised you did not find anything. Here's what a quick search through Google Scholar with keywords path stroke recognition and handwriting stroke recognition got me:

One more thing to add: The most recent reference in the above bunch is as old as 2000. That indicates that what you want to do is probably a well-researched area, with tried out state-of-the-art techniques. It should not have been hard to look for reference material.

Just to clarify, I did not actually read any of these articles. I just did my usual preliminary research step: go to article databases (google scholar, mendeley, ieeexplore...), type in a few keywords. Open the ones with promising titles, close one with completely uninteresting abstracts. Read the remaining abstracts more carefully, and select the ones that seem relevant. Having said this, not all the articles I proposed might be relevant, but judging from experience, they most probably are.

I hope this helps. Also, if you have done any research in the meantime, please edit it in your question and share with the community. It looks like an interesting topic.

Community
  • 1
penelope
  • 3,676
  • 1
  • 26
  • 63
0

You should use a neural network, that's what is used for optical character recognition, I'm sure you can find a lot of ressources about it on Google. To take into account the order of the lines, you can use not just the final image but also the intermediate images after each stroke.