Catálogo de publicaciones - libros

The theoretical framework of Statistical Learning Theory (SLT) for pattern recognition problems is extended to comprehend the situations where an infinite value of the loss function is employed to prevent misclassifications in specific regions with high reliability.

Sufficient conditions for ensuring the consistency of the Empirical Risk Minimization (ERM) criterion are then established and an explicit bound, in terms of the VC dimension of the class of decision functions employed to solve the problem, is derived.

In this paper, we investigate the full equivalence, under basic conditions, between the Probabilistic PCA clustering approach and the reconstruction of signal subspaces based on the singular value decomposition. Therefore this equivalence allows the adaptive determination of the clusters identified on data, in order to maximize the quality of the reconstructed signal. Furthermore, using known results in SVD framework, we also introduce a new technique to estimate automatically the dimension of the latent variable subspace.

Dominant setsare a new graph-theoretic concept that has proven to be relevant in pairwise data clustering problems. We address the problem of grouping out-of-sample examples after the clustering process has taken place. This may serve either to drastically reduce the computational burden associated to the processing of very large data sets, or to efficiently deal with dynamic situations whereby data sets need to be updated continually. Numerical experiments show the effectiveness of the approach.

One way of using the entropy criteria in learning systems is to minimize the entropy of the error between two variables: typically, one is the output of the learning system and the other is the target. This framework has been used for regression. In this paper we show how to use the minimization of the entropy of the error for classification.

The minimization of the entropy of the error implies a constant value for the errors. This, in general, does not imply that the value of the errors is zero. In regression, this problem is solved by making a shift of the final result such that it’s average equals the average value of the desired target. We prove that, under mild conditions, this algorithm, when used in a classification problem, makes the error converge to zero and can thus be used in classification.

Detecting regions of change in multiple images of the same scene taken at different times is of widespread interest due to a large number of applications in diverse disciplines, including remote sensing, surveillance, medical diagnosis and treatment, civil infrastructure, and underwater sensing.

The paper proposes a data dependent change detection approach based on textural features extracted by the Independent Component Analysis (ICA) model. The properties of ICA allow to create energy features for computing multispectral and multitemporal difference images to be classified. Our experiments on remote sensing images show that the proposed method can efficiently and effectively classify temporal discontinuities corresponding to changed areas over the observed scenes.

In the last years Independent Component Analysis (ICA) has been applied with success in signal processing and many algorithms have been developed in order to perform ICA. In this paper we review some algorithms, like INFOMAX (Bell and Sejnowski 1995), extended-INFOMAX (Lee, Girolami and Sejniowski 1997), FastICA (OjA, and Hyvärinen 1999), that solve the ICA problem under the assumption of the linear mixture model. We also show an overview of the nonlinear ICA algorithms and we discuss the MISEP (Almeida 2003). In order to test the performances of the reviewed algorithms, we present some applications of ICA in biomedical signal processing. In particular the application of ICA to the electroencephalographic (EEG) and surface electromyographic (sEMG) recordings are shown.

Research on human sensorimotor functions has hugely increased after electromyogram (EMG) analysis was replaced by functional magnetic resonance imaging (fMRI), that allows to obtain a direct visualization of the brain areas involved in motor control. Very meaningful results could be obtained if the two analysis could be correlated. Our goal is to acquire the EMG data during an fMRI task. The main problems in doing this are related to the electromagnetic compatibility between the resonance coils (very high magnetic fields) and the EMG electrodes. In this study we developed a system that can characterize the entire EMG signal corrupted by the magnetic fields generated by the magnetic resonance gradients. The entire system consists in a hardware equipment (shielded cables and wires) and a software analysis (effective mean analysis and wavelet analysis). The results show that a motor task was correctly delivered by our post processing analysis of the signal.

Automatic object positioning in 3D space is nowadays required by a great variety of applications. We propose here a new approach to this problem, whose core is constituted by a bank of neural networks; from the measured positions of a set of laser spots generated on the object surface, the nets estimate the position of a set of points rigidly connected to the object. Results on synthetic data are reported, and show that the proposed method is reliable and comparable in accuracy with the most common solutions present in the literature, which are based on Iterative Closest Point (ICP) matching.

Pixel failures often introduce in digital images a characteristic impulsive noise, known as “salt & pepper”. This has to be corrected to get clear digital images. In this paper a new approach to the problem, based on an adequate model of the sensor and on the properties of the Human Visual System, is introduced. The local background luminance is estimated through a 3x3 median filter, and noise standard deviation from the sensor model. Since the filter is based only on local operations, it can work at real-time rates (less than 0.7s for 12 bit, 4.8MPixel images). Its speed may be even improved by using DSP implementation.

The present work is part of a wider research activity carried on within the Italian National Project named SIINDA. It shows how physical atmosphere parameters like temperature, humidity, wind direction, can be indirectly estimated in specific points of the monument, if one, or more than one, ambient air monitoring station is present in the neighborhood of the monument itself. We use a connectionist system trained to map the parameters measured by such stations with the parameters measured by the set of installed sensors. The obtained results look like very good and we received the approving by cultural heritage experts who evaluated such a methodology to effective by support monitoring in the field of the conservation state of monuments.