Tweet
Dear all,
I am collaborating on a project involving analysis of hand writing data. Some of the data are acquired using a Tablet PC (Thinkpad X61t), and – of course - pixelated. I wonder what is the best method to preprocess these data in order to minimize artifacts due to pixelation and unsystematic error/noise and would appreciate your support in choosing a good approach.
Some of the methods that have been used are
- Gaussian smoothing
- low pass filter (Butterworth vs. FIR?)
- Savitzky-Golay smoothing
- non-parametric regression methods
The last two seem to be particularly attractive as they allow to compute not only smoothened data but also higher derivatives. I think I understand the Savitzky-Golay approach (local polynomial fit), but don’t understand non-parametric regression methods, as discussed for instance in Marquardt & Mai (1994), A computational procedure for movement analysis in handwriting.
I’d be grateful for recommendations and general information (if possible with references), as well as specific information concerning choice of parameters (cut-off frequency, window size) and implementation of the non-parametric approach.
Best regards,
Julius Verrel
Center for Lifespan Psychology
MPI for Human Development, Berlin, Germany
I am collaborating on a project involving analysis of hand writing data. Some of the data are acquired using a Tablet PC (Thinkpad X61t), and – of course - pixelated. I wonder what is the best method to preprocess these data in order to minimize artifacts due to pixelation and unsystematic error/noise and would appreciate your support in choosing a good approach.
Some of the methods that have been used are
- Gaussian smoothing
- low pass filter (Butterworth vs. FIR?)
- Savitzky-Golay smoothing
- non-parametric regression methods
The last two seem to be particularly attractive as they allow to compute not only smoothened data but also higher derivatives. I think I understand the Savitzky-Golay approach (local polynomial fit), but don’t understand non-parametric regression methods, as discussed for instance in Marquardt & Mai (1994), A computational procedure for movement analysis in handwriting.
I’d be grateful for recommendations and general information (if possible with references), as well as specific information concerning choice of parameters (cut-off frequency, window size) and implementation of the non-parametric approach.
Best regards,
Julius Verrel
Center for Lifespan Psychology
MPI for Human Development, Berlin, Germany