Compression of hyperspectral images using block coordinate descent search and compressed sensing
Shirin Hassanzadeh, Azam Karami, Rob Heylen, Paul Scheunders · 2016
In this paper, a new lossy compression method for hyperspectral images (HSI) is introduced based on the NonNegative Tucker Decomposition (NTD). HSI are considered as a 3D dataset: two spatial dimensions and one spectral dimension. The NTD algorithm decomposes the original data into a smaller 3D dataset (core tensor) and three matrices. In the proposed method, in order to find the optimal decomposition, the Block Coordinate Descent (BCD) method is used, which is initialized by using Compressed Sensing (CS). The obtained optimal core tensor and matrices are coded by applying arithmetic coding and finally the compressed dataset is transmitted. The proposed method is applied to real datasets. Our experimental results show that, in comparison with state-of-the-art lossy compression methods, the proposed method achieves the highest signal to noise ratio (SNR) at any desired compression ratio (CR) while noise reduction is simultaneously obtained.