Wavelet Kernel Principal Component Analysis in Noisy Multiscale Data Classification

Figure 5

The classification accuracy with respect to different numbers of principal components retained under Gaussian and Mexican hat TI wavelet kernels, using the coefficients of wavelet approximation at various decomposition level. The classifiers used are FLD and 1-NN.

(a) Gaussian kernel, at level 10

(b) Mexican hat kernel, at level 10

(c) Gaussian kernel, at level 9

(d) Mexican hat kernel, at level 9

(e) Gaussian kernel, at level 8

(f) Mexican hat kernel, at level 8