(a) (b) (c)

he impact of ߣ, which is a smoothing parameter, on the baseline estimation

nd the signal discovery performance. (a) ߣൌ0.6. (b) ߣൌ1.1. (c) ߣൌ2.

stance, Figure 5.3 shows three models using the spline approach

ate a baseline for the same spectrum using the waveslim

in which a smoothing parameter was tuned for the discovery of

baseline. Three smoothing parameter values were used for the

ation of the impact of a smoothing parameter on the smoothing

nce. Because of the use of different smoothing parameters, the

were different. Figure 5.3(a) shows an over-estimated baseline

the regularisation constant was too small. In this situation, the

nimisation was strengthened while the degree of the smoothness

kened. The consequence was that the peaks or signals were all

into the estimated baseline, hence the signals were hardly

Figure 5.3(c) shows an under-estimated baseline. When the

ation constant was too large, the smoothness was strengthened

error minimisation was weakened. The consequence was that the

was a straight line and some true peaks were missed and some

ks were present. Figure 5.3(b) shows a well-estimated baseline,

e regularisation constant was more properly selected. In this

and only in this situation, the majority of the true signals were

ed and the chance of the presence of the false signals was

d.