This tutorial will quickly demonstrate how to refine and add new characters to Recog's repetoir. Before starting, choose the Supervised option in the Learning menu.
Open the character net file NUMBERS.RCG. This network will recognize mouse-drawn digits from 0 to 9. Naturally it was trained using my handwriting. I draw my 1s with large serifs on the top and bottom. Try drawing a 1 as a simple vertical line, then digitize and recognize it. It will probably rank the 1 at around 50% since a vertical bar doesn't match my version.
To correct this, select the correct digit from the list (in this case the 1), and then click the Learn button. When the Learn button is used, Recog modifies the neural net of the selected digit so that it more closely matches the pattern currently displayed on the grid of pixels.
Repeat the process twice more (redraw the simple 1, digitize it and learn it). Each time it should get better at recognizing that what you are drawing is a 1. It is slowly forgetting about my handwriting and learning yours.
Draw an X, digitize it and recognize it. The low percentages should indicate that it hasn't any idea what that character is. Click the Add button and enter X as the name of the new character. This will create a blank character net; it hasn't done any learning yet. Choose the X from the list and click Learn. Feed it two more examples X (learning each one) and it should have a pretty good idea of how to recognize this new character.
This style of learning is called Supervised since you had to tell it every step of the way when to learn and when to add characters. The next tutorial describes how to use Unsupervised learning.