Two stage quantization of noisy hyperspectral images

A two-stage quantization approach for compression of noisy hyperspectral images is proposed. In the first stage, a multilevel denoising process uses the minimum noiseless description length (MNDL) approach to not only denoise the data, but also provide quantization levels for the noise dominant wavelet coefficients. In the second stage, the remaining noiseless dominant coefficients are quantized with the conventional quantization methods such as the high bit rate uniform quantization approach. Our simulation results show the advantages of the proposed method over separate “denoising and compression” approaches in both improving the output SNR and providing much less number of quantization levels.