LAboratoire de Spectrochimie Infrarouge et Raman – UMR 8516
random_5
  • Français
  • English
  • Spanish

Séminaire de Paul Eilers

Erasmus University Medical Centre, Rotterdam, The Netherlands
The power of penalties in signal processing

It seldom occurs that the signals we measure with our instruments are perfect. There can be noise, a drifting baseline, artefacts or unwanted filtering. By proper pre-processing we can reduce their influence. It is useful to look at it as the estimation of latent signals with certain properties. An example is a trend: a smooth curve that goes through the middle of a series of noise measurements. Another example is a baseline: a (very) smooth curve, following the bases of a series of peaks.

In a number of cases we can translate the desired properties of a latent signal into a so-called penalty, a number that is small when the signal behaves the way we like, and large when it does not. In a statistical model, that describes the relationship between the latent signal and the observations, the penalty can be introduced to steer the solution in a desired direction.

I will present several examples of the use of penalties, avoiding unnecessary mathematical details. They are all related to the cooperation between our institutes in the field of spectroscopy and image analysis.

This is joint work with Johan de Rooi.

Le séminaire aura lieu le 12.04.2012 à 10h30
salle de réunion du LASIR, sous-sol bâtiment C5