Request a Personalized Demo

Learn how to accurately convert your designs with Scan2CAD

  • Join thousands of happy customers worldwide

How to train Scan2CAD to recognize a font


To train Scan2CAD to recognize a font you need to follow these steps:


1. Create a new training set

A training set is a set of example characters from which a neural network learns to recognize a font.

Create a new training set using Train Menu > Initialize Training Set.


2. Open a raster image containing example fonts

You normally add example characters to the training set by selecting them from a raster image or images.

Click to open a raster image that contains examples of the font you want to train the neural network to recognize.


3. Thin the raster image

Thin the raster image using Train Menu > Thin Image.


4. Add example characters to the training set

There are several methods for adding example characters to the training set. See Add characters to a training set.

You can use just one method or combine all methods.

You can extract example characters from any number of raster images and add them to the training set at any time. You do not have to build the whole training set from one raster image in one session. Use Train Menu > Load Training Set to load an existing training set so you can continue to add example characters to it.


5. Save the training set

Save the training set using Train Menu > Save Training Set As.


6. Initialize a neural network

Create a new neural network using Train Menu > Initialize Neural Net. A dialog appears. This contains two options: Auto and Advanced.

Under normal circumstances you should initialize the neural network automatically using the Auto option.


7. Train the neural network

Train the neural network using Train Menu > Train Neural Net.


8. Save the neural network

Save the neural network using Train Menu > Save Neural Net As.


9. Test the neural network

Click to load a raster image that you want to test the newly trained neural network on, then select Train Menu > Test Neural Net.

Scan2CAD performs OCR on the raster image using the currently loaded neural network.


If the test shows that the neural network is consistently failing to recognize a character or characters, you may be able to improve performance by adding more example characters to the training set or by training the neural network for longer. You should also check that the character has been correctly assigned and that it is adequately represented in the training set. See Troubleshooting poor neural network performance.


10. Use the neural network

To use a neural network you have trained, go to OCR Menu > Settings and select the User Character Set in the OCR Settings dialog.

Select the neural network you want to use from the drop down list. You should also make sure that Display Mode is set to Strings.

Next time you use convert text using Run Menu > OCR () or Run Menu > All () Scan2CAD will use the selected neural network.


Have questions on this topic? Talk to us