I am trying to create a dataset where the Text is mixed, e.g. " I love football" can be written as " I l()v3 F00tba11". The idea is that by using Tesseract I can find the pattern to match these two types of text images and normalize the text.
So which should be the best approach. Supervised or Unsupervised? How many images do I need for each for a sample result. If it is supervised, how should I label them? I am very new to data science so it is confusing for me, so please suggest the best approach to this.
If anyone knows some kind of dataset fitting my example is really welcome.