Abstract:

In this paper, we propose a novel algorithm for hyperspectral image (HSI) denoising which is based on the tensor subspace. Considering the HSI as 3 order tensor data, our approach introduce a data representation involving multidimensional processing and project such data into the signal subspace by tensor subspace decomposition. The optimization criterion used in this algorithm is the minimization of mean square error between the estimated signal and the desired signal, then the alternating projection algorithm is adopted to determine the optimal filter in each dimension. Comparative studies with conventional denoising methods such as 2-D Wiener filtering and channel-by-channel wavelet thresholding show that our algorithm provides better performance using AVIRIS and PHI datasets.

Key words: hyperspectral image, denoising, tensor subspace, eigenvalue decomposition