The best way to get started is to download or clone this repository and play around with the example scripts in the startup folder. This walkthrough is very high level and is meant to accompany those example scripts.
Please note: Certain features (such as downloading a model or loading an image) do not work over the file:// protocol for local development. You need to use http:// or https:// to load the page or you'll get a lot of CORS errors in your browser console. An easy way to get a local HTTP server up and running is to use the http-server tool.
Keep your developer console open as this library outputs a lot of very useful debug information there.
Create a copy of train.html from the starter folder, update the CBL initialization settings, and add links to images from your chosen CAPTCHA.
Most of your time will be spent adjusting the image manipulation methods in the preprocess method so that we can clean up the image enough to extract individual characters.
Please note: The very last step of the preprocess method needs to include a call to colorRegions or the segmentation will not work.
Once you are happy with the quality of the segmentation, call the train method to bring up a dialog that will walk you through categorizing and training.
After training, save the model as model.txt.
Create a copy of solve.html from the starter folder and make sure your model.txt from training is in that same folder.
Also make sure you're using the exact same settings when instantiating the CBL object as you did when training the model.
