⛔️ This article is for a legacy version of Scan2CAD. See our updated tutorials. ⛔️


Thin the raster image


Thin Image in the Train Menu is exactly the same as Thin > Standard in the Raster Effects Menu. It is also included in the Train Menu because raster images must be thinned before characters on them are added to a training set.

This is because when an example character is added to the training set Scan2CAD “standardizes” it by scaling it to fit within a grid of 12 x 16 pixels. This allows the example characters to be presented to the neural network in a consistent manner for learning.

Thinning the characters means that shape details are retained when the characters are scaled. In the diagrams below, the “S” on the left was not thinned before scaling and is no longer recognizable as an “S”. The “S” on the right is the same “S” scaled after thinning.



1. Select Train Menu > Thin Image.
2. In the Preview Window, zoom into the image so that you will be able to clearly see the effects of the thinning.
Warning: If you display a large section of your raster image in the Preview Window this function may operate very slowly.
3. Select a number of thinnings by moving the slider in the dialog.

The number of thinnings is the number of times Scan2CAD goes through the image stripping a layer of pixels off it. The larger the number you choose, the more thinned the image will be although it will never be thinned right away – Scan2CAD will always retain an image one pixel thick.

As you move the slider to the right you will see the raster image in the Preview Window getting thinner. For the purposes of font training, the number of thinnings you enter should be large enough to thin the image to one pixel thickness.

4. Zoom and pan around the image in the Preview Window to make sure the number of thinnings you have chosen is appropriate to other parts of the image.

Adjust the number of thinnings if necessary.

5. Ensure the Apply changes to Full Image option is selected.
6. Click OK.


Click to undo.


If a Pick Color is selected, Scan2CAD will only thin Pick Colored parts of the raster image.
If you are dealing with characters that display extreme variations in size, it is advisable to train separate neural networks to recognize the large and small characters. This is because characters of vastly different sizes may not scale consistently when added to the training set.


Have questions on this topic? Talk to us