I am currently writing my bachelors thesis about on-line handwriting recognition.
In this context, there are different methods to make the input data more useful to classifiers / to make learning more effective:
- Scaling the drawn symbol
- Shifting the drawn symbol
- Wild point detection
I think all of those techniques are preprocessing, because I apply them before the image is "processed" by the training / recognition algorithm.
But aren't all of those techniques also normalization techniques? Where is the difference between "preprocessing" and "normalization"?