Adaptive multi-way analysis of images

This paper presents a new multi-way filtering method for multidimensional images corrupted by white Gaussian noise. Images are considered as multi-way arrays instead of matrices or vectors, which enables to keep relations between each index. The presented filtering method is based on multilinear algebra principles and it improves the multi-way Wiener filtering (MWF). The originality of the method relies on the flattening directions of multi-way arrays and on a block approach to keep local characteristics of images. Experiments on color images and hyperspectral images have been computed to illustrate the improvement of MWF by the analysis of image characteristics.