Abstract
An improved method of wavelet threshold denoising is introduced and applied to hyperspectral image denoising. The method estimates a threshold value for each spectrum. When the signal is transformed to the wavelet domain, a large number of coefficients with small (or zero) values and a small number of coefficients with large values are gotten. However, transforming the noise to the wavelet domain produces sort of a scattered distribution of the noise energies over all scales and translations, assuming that the noise is white. By completing these wavelet transforms, the original data cube will be completely transformed to the wavelet space. Each component in the transformed cube is considered as a wavelet coefficient that represents the frequency distribution of the input data, which can reduce the noise in the spectral domain at the same time. Thresholds are set to a scalar specifying the percentage of cumulative power to retain in the filtered wavelet transform. Find the actual percent corresponding to these coefficients. During the processing, four families of mother wavelets (Symlets, Daubechies, Haar and Coiflet) are tested in a series of experiments to estimate the functioning of those wavelets and threshold parameters. Experimental results show that the proposed algorithm with Symlets provides an improvement in SNR for hyperspectral data specially.
Original language | English |
---|---|
Article number | 71564D |
Journal | Proceedings of SPIE - The International Society for Optical Engineering |
Volume | 7156 |
DOIs | |
Publication status | Published - 2009 |
Event | 2008 International Conference on Optical Instruments and Technology: Optical Systems and Optoelectronic Instruments - Beijing, China Duration: 16 Nov 2008 → 19 Nov 2008 |
Keywords
- Hyperspectral data
- Wavelet Threshold Denoising