Razor PIC - Find peaks in noise. Detect peaks that are
obscured. Measure peak positions and peak widths for peak
Using a Bayesian second derivative, we have developed
a peak picker that takes proper account of the noise, finds peak locations
and estimates their height and width. It locates occluded peaks that the eye
cannot discern, and estimates their statistical significance, reporting the
results as signal to noise ratios based either on amplitude or area.
Fig 1 shows its performance on a Raman spectrum
(sulfur). The peak model was chosen to be a Lorentzian peak of the same
width as the dominant peak in the spectrum. The height of each marker is
proportional to the peak significance. The picker was set to pick by
Fig 2 shows a typical problem. We intend to fit this
spectrum with model peaks, but to do so we must know how many and where they
are. We use the narrowest peak in the spectrum to set the picking
resolution, and decide to accept every peak whose amplitude significance
> 3 standard deviations of the noise.
The peak file that is written for FIT contains the
estimated width and height of each peak. There is little else to do but
invoke FIT, and sit back. Everything is automatic; no need to drag peaks
around and fuss with the peak parameters.