An interscale multivariate map estimation of multispectral images

In this paper, we develop a multivariate statistical approach for image denoising in the wavelet transformed domain. To this respect, the wavelet coefficients of all the image channels at the same spatial position, in a given orientation and at the same resolution level, are grouped into a vector and a maximum a posteriori estimate is derived from a multivariate Bernouilli-Gaussian prior. The parameters of this statistical model are computed recursively from coarse to fine resolutions in order to exploit the inter-scale redundancies between the wavelet coefficients. Simulation tests performed on remote sensing multispectral images indicate that the proposed procedure improves the conventional wavelet-based denoising methods.