Methods coming from statistics and pattern recognition to estimate the cloud mask from radiance measured by visible and infrared sensors on board satellites are gaining greater consideration for their ability to properly exploit the increasing number of channels available with current and next-generation sensors. Endowed with physical arguments, they give rise to robust methods for accurately estimating the cloud mask. Application of such classification methods to Moderate Resolution Imaging Spectroradiometer (MODIS) data is discussed in this paper. Three different types of MODIS datasets are considered: synthetic (radiance is simulated by proper radiative transfer models); annotated (real MODIS data labeled by a meteorologist as clear or cloudy); and real MODIS data, whose truth is obtained from the official MODIS cloud mask product. A full assessment of the MODIS spectral bands is performed, aimed at understanding the role of the spectral bands in detecting clouds and at achieving top performance with very few properly chosen spectral channels. Local methods that use spatial correlation of images to improve classification, reducing the pseudonuisance of nonlocal methods, have also been tested on real data.

This work investigates the capability of supervised classification methods in detecting both major tissues and subcortical structures using multispectral brain magnetic resonance images. First, by means of a realistic digital brain phantom, we investigated the classification performance of various Discriminant Analysis methods, K-Nearest Neighbor and Support Vector Machine. Then, using phantom and real data, we quantitatively assessed the benefits of integrating anatomical information in the classification, in the form of voxels coordinates as additional features to the intensities or tissue probabilistic atlases as priors. In addition we tested the effect of spatial correlations between neighbouring voxels and image denoising. For each brain tissue we measured the classification performance in terms of global agreement percentage, false positive and false negative rates and kappa coefficient. The effectiveness of integrating spatial information or a tissue probabilistic atlas has been demonstrated for the aim of accurately classifying brain magnetic resonance images.

Clustering is one of the most important unsupervised learning problems and it consists of finding a common structure in a collection of unlabeled data. However, due to the ill-posed nature of the problem, different runs of the same clustering algorithm applied to the same data-set usually produce different solutions. In this scenario choosing a single solution is quite arbitrary. On the other hand, in many applications the problem of multiple solutions becomes intractable, hence it is often more desirable to provide a limited group of ''good'' clusterings rather than a single solution. In the present paper we propose the least squares consensus clustering. This technique allows to extrapolate a small number of different clustering solutions from an initial (large) ensemble obtained by applying any clustering algorithm to a given data-set. We also define a measure of quality and present a graphical visualization of each consensus clustering to make immediately interpretable the strength of the consensus. We have carried out several numerical experiments both on synthetic and real data-sets to illustrate the proposed methodology.