Improved classification and segmentation of hyperspectral images using spectral warping
International Journal of Remote Sensing
| Hi-index | 0.00 |
Spectral warping is a time domain to time domaintransformation on a signal that effectively warps thefrequency content of the original signal. Here we presenta matrix formulation of the spectral warpingtransformation. The transform matrix is decomposed intothree steps. The first is a DFT to convert the time signalinto the frequency domain. Step two is an interpolationmatrix to calculate the signal content at the desired newfrequency samples. This effectively provides the frequencywarping. The final step is an inverse DFT to transformthe signal back into the time domain.A direct consequence of this matrix representation is adirect FIR implementation of spectral warping, ratherthan the more commonly used IIR technique. Wedemonstrate that spectral warping is a generalisation oflinear filtering, and show how the conventional all-passspectral warping transformation can be generalised byusing either arbitrary frequency mapping functions ordifferent interpolation schemes. Finally, the conditionsfor the invertibility of the spectral warping transformationare derived.