5
$\begingroup$

Excuse any mistakes in my description as I'm new to ML. But I have an application that takes user input to generate paths/curves (All symbols are single paths) and I would then like to attempt identification. This seems, generally, to be a pretty well studied problem, and there is a lot of reference I can find.

However everything I've found so far starts with some sort of raster format, which makes sense given that often recognition comes in initially in that format. But given my particular set of constraints it seems that there may be additional useful data given that I have paths.

So my question is are there any good techniques for doing identification without first rasterizing my paths that may be well suited for my particular instance, or is rasterizing to a grid particularly well suited to the problem, and should I just raster and solve this problem more classically?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
4
$\begingroup$

What you are looking for is called "on-line recognition". I have written my Bachelors thesis about this:

Thoma, Martin. "On-line Recognition of Handwritten Mathematical Symbols." arXiv preprint arXiv:1511.09030 (2015).

And I've created the online service write-math.com as a part of it.

As a short introduction, you might be interested in my presentations.

Are there any good techniques for doing identification without first rasterizing my paths that may be well suited for my particular instance?

Yes. You can on the one hand make it a fixed-length sequence, on the other hand you can use recurrent neural networks (I didn't try that in my bachelors thesis)

Is rasterizing to a grid particularly well suited to the problem, and should I just raster and solve this problem more classically?

This works astonishingly well. I achieved better results with the rasterized version (in my master thesis which isn't published yet due to bureaucracy problems at my university) than with fixed-length sequence techniques. But I guess the way to go is to do both.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thanks for the Info, does your thesis have a title I could search for in the future once it's actually published. $\endgroup$
    – Arelius
    Commented Jun 19, 2017 at 6:31
  • $\begingroup$ Specifically referring to your master's thesis. $\endgroup$
    – Arelius
    Commented Jun 19, 2017 at 6:33
  • $\begingroup$ "Analysis and Optimization of Convolutional Neural Network Architectures" - I can leave a comment on 3rd of August (then I'm allowed to publish it) $\endgroup$
    – Martin Thoma
    Commented Jun 19, 2017 at 7:50
  • $\begingroup$ Great, thanks. Looking forward to reading it. $\endgroup$
    – Arelius
    Commented Jun 27, 2017 at 22:44

Not the answer you're looking for? Browse other questions tagged or ask your own question.