Exploiting Optimal Supports in Enhanced Multivariance Products Representation for Lossy Compression of Hyperspectral Images

Thanks to advancements in technology, the importance of computational methods used in tasks like storing and processing data is increasing as the data produced becomes more complex in both size and detail. Methods such as Tucker Decomposition, CANDECOMP/PARAFAC, Alternating Least Squares and their derivations, are widely used in the field to meet the requirements in numerous areas. These cases contain expressing high-dimensional data using lower-dimensional tensors, cleansing the data of errors that occur during data acquisition while also ensuring an efficient compression. This study proposes a new method that exploits the tensor structure of 3-dimensional data by calculating the lower-dimensional components via Enhanced Multivariance Products Representation and produces a superior approximation compared to well-known tensor decomposition methods. An iterative process is established to calculate the optimal support tensors and to determine the lower-dimensional components, which in further steps are employed to reconstruct the approximation.