|
|
Development of novel intrinsic dimension estimators:
This project is focused on the development of novel techniques for the automatic estimation of intrinsic dimension, which is generally defined as the minimum number of parameters needed to represent the input data without information loss. Alternatively, it is defined as the (topological) dimension of the underlying manifold, from which the dataset is supposed to be drawn. Although many estimation techniques have been proposed in the literature, most of them fail on noisy data, or compute underestimated values when the intrinsic dimension is sufficiently high. We are currently investigating the theoretical motivations of the bias that causes the underestimation effect; the acquired knowledge and the developed theories are exploited to develop estimators based on statistical properties of manifold neighborhoods, formalized as functions of the intrinsic dimension.
Face recognition
This project addresses the problem of face recognition under uncontrolled conditions. To this end we are exploring the potentiality of a sparse representation classifier based on suitable nonconvex parameter dependent mappings which provide an easy and fast iterative schema for the classification. Preliminar experimental results show the validity of this approach, encoraging further investigations.
Recognition of facial emotions
“The face is the most extraordinary communicator, capable of accurately signaling emotion in a bare blink of a second, capable of concealing emotion equally well” Deborah Blum
This project aims to recognize automatically the expressions and the emotions of people, by processing video sequences representing an expressive face. Computer vision techniques are adopted to extract the peculiar image features, and Bayesian networks are used to characterize the expression dynamic.