Scaling of the Data

Command: Preprocessing > Scaling the Data

ImageLab provides several methods to scale the data. The user can either scale spectral data, data within a layer, or the entire data set:

  • Spectra: each spectrum of a particular time slot is scaled according to the selected type of scaling
  • Layers: all data within one layer of a particular time slot are scaled using the selected type of scaling
  • Entire data set: all data of all time slots are scaled

 

Available types of scaling:

Mean centering The selected data items (layers, spectra, or all data) are scaled in such a way that the mean of each item becomes zero.
Standardisation The selected data items are scaled to a zero mean and a standard deviation of 1.0. Please note that the standardisation of data may be problematic if the data contains unusually large outliers. In this case q-normalisation is a better alternative.
Constant sum The selected items are scaled to a constant sum defined by the parameter A.
Constant sum of squares The selected items are scaled to a constant sum of squares. The sum is specified by the parameter A.
Maximum amplitude The selected items are scaled in such a way that the maximum absolute value of each item becomes A.
Range The selected data values are scaled to cover a range between A and B.
Q Normalisation The selected data range is scaled to zero median and a difference between the median and the q-percentile of 1.0, with q (in %) given by the parameter A. Q normalisation is largely insensitive to outliers and should be used whenever you are expecting severe outliers.
Squashing Function The selected data range is compressed by applying a sigmoid function ("squashing function") to the interval [-1,+1]. The parameter A specifies the origin (offset) of the squashing function, the parameter B defines the slope of the function.

 

How To:
  1. Select the scaling range (box 'Apply Scaling to')
  2. Select the type of scaling
  3. If required, specify the parameters for the selected scaling type
  4. Optionally select a pixel mask which restricts the scaling process to pixels which are not masked
  5. Select the spectra to be processed (if the data set contains more than one spectral group)
  6. Optionally restrict the range of the selected layers
  7. Click the "Apply" button

Hint: If you abort the scaling operation the data will be partially changed and thus unusable for further processing. After aborting the scaling you should reload the data set from disk.


Last Update: 2018-Sep-12