IMAGES FUSION BASED ON A COUPLED NONNEGATIVE TENSORS FACTORIZATION APPROACH (CNTF), APPLICATION TO OLCI AND ETM SENSORS

摘要

Coastal water monitoring often requires images with good spatial and spectral resolutions. This communication describes a new method for multispectral and hyperspectral images fusion: the Coupled Nonnegative Tensor Factorization fusion approach which directly exploits the tensorial nature of multi and hyperspectral images. The CNTF fusion algorithm allows to separate both hyperspectral and multispectral images into two matrices: one of endmembers and one of abundance. For application and validation purposes, a multispectral image (ETM) and a Sentinel 3/OLCI image was simulated from a Hyperspectral Imaging for Coastal Waters image (HICO). The resulting fusion image was compared with a reference image which was generated from the HICO image too. Statistical parameters such as RASE, ERGAS, SAM, correlation and bias, are computed for the spectral and spatial assessment of the fused products.