What are some current methods for implementing OCR on handwriting?
This is what I have so far:
* Pre-process with some image filters, try combinations of: Black and white, Greyscale, edge detection(Canny?), line detection(Hough?)
* Try kNN (k nearest neighbor) algorithm first. This should be fast and work on easy stuff, but its more of a math equation and not really deep learning.
* Try a CNN (convolutional neural network). Probably use either the Keras or mxnet libraries
STILL tripping me up:
* Is there another machine learning technique I should try to apply to this other than the kNN and CNN strategies??
* Are there libraries that are more specifically geared towards handwriting OCR than Keras or mxnet? RNNLIB (too old?), ocropy?
* Lastly, I haven't come across many methods to isolate characters and words from an image full of handwritten text. Figured I'd try to pull characters out and send them through the kNN algorithm, and pull characters and words out before putting them through the CNN engine.
It works great with unusual fonts/handwritten allcaps, and it's the absolute best when it comes to working with low-resolution images.
I have no experience using it with hard-to-read handwriting, but it might be worth checking it out if you don't know it already.