Charec is a minimalistic handwriting recognition program, written in less than 300 lines of JavaScript.
Demo: https://ceptord.net/20191127-charec.html
Charec uses Levenshtein distance to recognize handwriting. Here is how it works:
-
Charec captures the finger/mouse movement as a sorted array of XY coordinates.
-
It compresses the array using some heuristics into a short string. This string represents the gist of the gesture.
-
Then, it performs a nearest neighborhood search against the predefined models (using Levenshtein distance). It returns the model with the lowest score as the result.
For more details, please read the following note.
The popular view of Levenshtein distance is that you can use it to compute string similarity. If you input two strings, it will return a number that shows how alike they are - the smaller, the more similar. This function is useful for many linguistic tasks.
For instance, suppose you're trying to find the correct spelling of a mistyped word. You feel the spelling is not exactly correct, but not far off. In theory, you can iterate over pages of an English dictionary to find out the correct spelling, but it is rather tiresome to do so.
So instead, you ask computers to flip through an English dictionary. Since computers do not have an innate understanding of "similarity", you teach computers to use the Levenshtein distance, and tell it to return the word with the most lowest score. The basic point here is that Levenshtein distance works as a rough approximation of the human perception of string similarity, and computers can easily understand it.
So far, we talked nothing peculiar. Indeed, this is mostly how the Levenshtein distance is used in real world, and I'd say there is nothing particularly wrong with that view.
However, there is a more fun (albeit less common) view of Levenshtein distance: To put it short, we can think of it as a similarity metric for everything.
That is, it is a universal similarity metric that can measure the similarity of any objects (e.g. images, graphs or handwritings). The catch here is that you need to find some way to represent these objects as strings (so that a Levenshtein Distance function can work on them). If your string representation is reasonably good, Levenshtein distance can act as a similarity metric for these objects.
Here is how it typically goes like:
(1) Take an object X that you want to measure the similarity.
(2) Find some way to convert X into a sequence of characters.
(3) Measure the Levenshtein distance.
(4) Now you have similarity metrics for X.
This is essentially how charec works. It takes a set of strokes traced by a pointer, and compare their similarity by passing them to a Levenshtein distance function.
For encoding, I used the Schimke-Vielhauer-Dittman technique [1] and it turned out to work surprisingly well.
- Schimke, Sascha, Claus Vielhauer, and Jana Dittmann. "Using adapted levenshtein distance for on-line signature authentication." Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004.. Vol. 2. IEEE, 2004. https://zeus.fh-brandenburg.de/~vielhaue/jabreflib/[ScVD2004].pdf
Public Domain