F. Isgrò

Unsupervised entity and relation extraction from clinical records in Italian

This paper proposes and discusses the use of text mining techniques for the extraction of information from clinical records written in Italian. However, as it is very difficult and expensive to obtain annotated material for languages different from English, we only consider unsupervised approaches, where no annotated training set is necessary. We therefore propose a complete system that is structured in two steps. In the first one domain entities are extracted from the clinical records by means of a metathesaurus and standard natural language processing tools. The second step attempts to discover relations between the entity pairs extracted from the whole set of clinical records. For this last step we investigate the performance of unsupervised methods such as clustering in the space of entity pairs, represented by an ad hoc feature vector. The resulting clusters are then automatically labelled by using the most significant features. The system has been tested on a fairly large data set of clinical records in Italian, investigating the variation in the performance adopting different similarity measures in the feature space. The results of our experiments show that the unsupervised approach proposed is promising and well suited for a semi-automatic labelling of the extracted relations.

Semantic Cluster Labeling for Medical Relations

In the context of the extraction of the semantic contents important for the effective exploitation of the documents which are now made available by medical information systems, we consider the processing of relations connecting named entities and propose an unsupervised approach to their recognition and labeling. The approach is applied to an Italian data set of medical reports, and interesting results are presented and discussed from a qualitative point of view.

Integrating a Priori Probabilistic Knowledge into Classification for Image Description

This paper discusses a possible implementation of the integration of knowledge from a probabilistic ontology in the automatic description of images. This combination not only provides the relations existing between the different segments, but also improve the classification accuracy, as the context often gives cues suggesting the correct class of the segment.

A machine learning approach for predictive maintenance for mobile phones service providers

The problem of predictive maintenance is a very crucial one for every technological company. This is particularly true for mobile phones service providers, as mobile phone networks require continuous monitoring. The ability of previewing malfunctions is crucial to reduce maintenance costs and loss of customers. In this paper we describe a preliminary study in predicting failures in a mobile phones networks based on the analysis of real data. A ridge regression classifier has been adopted as machine learning engine, and interesting and promising conclusion were drawn from the experimental data.

Face Recognition from Robust SIFT Matching

This paper presents a face recognition algorithm based on the matching of local features extracted from face images, namely SIFT. Some of the earlier approaches based on SIFT matching are sensitive to registration errors and usually rely on a very good initial alignment and illumination of the faces to be recognised. The method is based on a new image matching strategy between face images, that first establishes correspondences between feature points, and then uses the number of correct correspondences, together with the total number of matches and detected features, to determine the likelihood of the similarity between the face images.

Improving Face Recognition in Low Quality Video Sequences: Single Frame vs Multi-frame Super-Resolution

Re-Identification aims to detect the presence of a subject spotted in one video in other videos. Traditional methods use information extracted from single frames like color, clothes, etc. A sequence in time domain of consecutive subject images could contain a greater amount of information compared with a single image of the same subject. Typically, these sequences are taken from surveillance cameras at very poor resolution. Even with modern cameras the resolution can be a problem when dealing with a subject who is far from the camera. A possible way of handling low resolution images is by using a multi-frame super-resolution algorithm. Multi-frame super-resolution image reconstruction aims at obtaining a high-resolution image by fusing a set of low-resolution images. Low-resolution images are usually subject to some degradation which causes substantial information loss. Therefore, contiguous images in a sequence could be viewed as a degraded version (SR image) of an image at higher resolution (HR image). Using a multi-frame SR algorithm could achieve a restoration of the HR image. This work aims to investigate the possibility of using a multi-frame super-resolution algorithm to enhance the performance of a classic re-identification system by exploiting information provided by video sequences made available by a video surveillance system. In the case that the SR technique employed results in an effective performance enhancement, we intend to show empirically how many match frames are required to have an effective improvement.

Exploiting Context Information for Image Description

Integrating ontological knowledge is a promising research direction to improve automatic image description. In particular, when probabilistic ontologies are available, the corresponding probabilities could be combined with the probabilities produced by a multi-class classifier applied to different parts in an image. This combination not only provides the relations existing between the different segments, but can also improve the classification accuracy. In fact, the context often gives cues suggesting the correct class of the segment. This paper discusses a possible implementation of this integration, and the first experimental results shows its effectiveness when the classifier accuracy is relatively low. For the assessment of the performance we constructed a simulated classifier which allows the a priori decision of its performance with a sufficient precision.

A Multi-spectral Stereo Method to Retrieve Cloud top Height applied to Geostationary Satellite images

In this paper we present a method to retrieve the Cloud Top Height (CTH), that is a refined version of a stereoscopic method present in literature. It is applied to stereo image pairs obtained by observations of the Meteosat Second Generation (MSG) geostationary satellites in a stereo setup. The performance of the method is tested both as mono band and multi spectral stereo. The estimated CTH are compared with the cloud altitude maps provided by the MODerate resolution Imaging Spectroradiometer (MODIS) on Terra-Aqua polar satellites. The results show that the new version of the method, performs better in comparison with the original algorithm despite the not proper stereo features of the system.

An experimental comparison of standard stereo matching algorithms applied to cloud top height estimation from satellite IR images

The JEM-EUSO (Japanese Experiment Module-Extreme Universe Space Observatory) telescope will measure Ultra High Energy Cosmic Ray properties by detecting the UV fluorescent light generated in the interaction between cosmic rays and the atmosphere. Cloud information is crucial for a proper interpretation of these data. The problem of recovering the cloud-top height from satellite images in infrared has struck some attention over the last few decades, as a valuable tool for the atmospheric monitoring. A number of radiative methods do exist, like C02 slicing and Split Window algorithms, using one or more infrared bands. A different way to tackle the problem is, when possible, to exploit the availability of multiple views, and recover the cloud top height through stereo imaging and triangulation. A crucial step in the 3D reconstruction is the process that attempts to match a characteristic point or features selected in one image, with one of those detected in the second image. In this article the performance of a group matching algorithms that include both area-based and global techniques, has been tested. They are applied to stereo pairs of satellite IR images with the final aim of evaluating the cloud top height. Cloudy images from SEVIRI on the geostationary Meteosat Second Generation 9 and 10 (MSG-2, MSG-3) have been selected. After having applied to the cloudy scenes the algorithms for stereo matching, the outcoming maps of disparity are transformed in depth maps according to the geometry of the reference data system. As ground truth we have used the height maps provided by the database of MODIS (Moderate Resolution Imaging Spectroradiometer) on-board Terra/Aqua polar satellites, that contains images quasi-synchronous to the imaging provided by MSG.

Comparing different methods to retrieve cloud top height from Meteosat satellite data

Cloud parameters such as the Cloud Top Height (CTH), Cloud Top Temperature (CTT), emissivity, particle size and optical depth have always been matter of interest for the atmospheric community. Particularly the CTH provides information leading to better understand the cloud radiative effects. Although there are many meteorological satellites providing the CTH, there are other sensors, not devoted to this purpose, that give some information from which this crucial parameter can be estimated. In this contribution we will describe three different methodologies to retrieve the CTH. The first technique is based on stereo-vision algorithms and requires two different views of the same scene and does not need of extra atmospheric information. In the second one, brightness temperatures in two IR spectral bands are converted to real cloud temperature by means of the proposed algorithms. From the CTT, the CTH is estimated using temperature vertical profiles (measured or modeled). The third technique retrieves the CTH from the output parameters of post event simulations performed by a Numerical Weather Prediction (NWP) model that in this work will be the mesoscale model WRF (Weather Research Forecast). This article presents a preliminary work, in which the heights retrieved by the three methodologies applied to the geostationary satellite Meteosat 10 are compared with the heights given by MODIS sensor installed on the polar satellite AQUA. This promising results show that valuable information about CTH can be retrieved from Meteosat which provide high frequency and large scale data useful for weather and climate research.