Using Image Processing For NLP?

I am wondering if it is feasible to perform some level of NLP by doing image processing. The idea is to create a diagram of the sentence in image form like this

Then, use an image classifier to train a model on a vocabulary of a user.

I am thinking of this in terms of the Kaggle Spooky Author Identification challenge.

I found this library:

I can create sentence diagrams, and turn them into images like this:

I wonder what will happen if I do this for a the whole vocabulary?

Seems like this would be more generalized if the diagram didn’t ave the actual text though.