Vincenzo Di Lecce

A Multi-expert System for Dynamic Signature Verification

This paper presents a multi-expert system for dynamic signature verification. The system combines three experts whose complementar behaviour is achieved by using both different features and verification strategies. The first expert uses shape-based features and performs signature verification by a wholistic analysis. The second and third expert uses speedbased features and performs signature verification by a regional analysis. Finally, the verification responses provided by the three experts are combined by majority voting.

An Evaluation of the Effectiveness of Image Features for Image Retrieval

Retrieval effectiveness in image databases depends significantly on the features and the distance model utilized to evaluate the similarity of the images. Features must be extracted from images and stored in the database. Since the features should be stored at the time of data entry, it is extremely important to determine which features ensure the best retrieval performances. In this paper a comparison of the most widespread automatic indexing techniques and their performances is presented. The image reference set, necessary for performance comparison, is obtained by including in the database frames extracted from video shots. Frames extracted from a shot are different but have the same semantic content. One of these frames is utilized as an example in a query; the indexing effectiveness is assessed from the frames retrieved. The most relevant features prove to be the angular spectrum, the Hough transform, and the color histogram, followed by local features such as local luminance pattern directionality.

Experimental System to Support Real-Time Driving Pattern Recognition

This work proposes an advanced driving information system that, using the acceleration signature provided by low cost sensors and a GPS receiver, infers information on the driving behaviour. The proposed system uses pattern matching to identify and classify driving styles. Sensor data are quantified in terms of fuzzy concepts on the driving style. The GPS positioning datum is used to recognize trajectory (rectilinear, curving) while the acceleration signature is bounded within the detected trajectory. Rules of inference are applied to the combination of the sensor outputs. The system is real-time and it is based on a low-cost embedded lightweight architecture which has been presented in a previous work.

FFT-based technique for image-signature generation

In this paper we address image retrieval by similarity in multimedia databases. We discuss the generation and use of signatures computed from image content. The proposed technique is not based on image annotation, therefore it does not require human assistance. Signatures abstract the directionality of image objects. They are computed from the image Fourier transform, and the influence of computation parameters on signature effectiveness is discussed. Retrieval is based on spectrum comparison between a reference image, assumed as the query, and the images in a collection. We introduce a metric for comparing the spectra and ranking the result, and approach the issue of partial query specification. Sample results on a small test collection are given.

Hierarchical-granularity holonic modelling

Design criteria for distributed and pervasive intelligent systems, such as Multi Agent Systems (MAS), are generally led by the functional decomposition of the given application-dependent knowledge. Consequently, changes either in the problem semantics or in the granularity level description may have a significant impact on the overall system re-engineering process. In order to tackle better these issues, a novel framework called Hierarchical-Granularity Holonic Model (HGHM) is introduced as a holon-based approach to distributed intelligent systems modelling. A holon is an agent endowed with special features. Seen from the outside, a holon behaves like an intelligent agent; seen from the inside, it appears to be decomposable into other holons. This property allows for modelling complex distributed systems at multiple hierarchical-granularity levels by exploiting the different abstraction layers at which the design process is carried out. The major benefit of the proposed approach against traditional holonic systems and MAS is that the entire HGHM-based architecture can be derived directly from the problem ontology as a hierarchical composition of self-similar, modular blocks. This helps designers focussing more on knowledge representation at different granularity levels which is a very basic process, as in top–down problem decomposition. Starting from the literature on holonic systems, a theoretical model of HGHM is introduced and an architectural model is derived accordingly. Finally, a customized application for the case study of distributed indoor air quality monitoring systems is commented and improvements in terms of system design with respect to well-established solutions are considered.

Computational-based volatile organic compounds discrimination: An experimental low-cost setup

In this work, an array of low-cost cross-sensitive sensors is used for discriminating the best candidate within a set of volatile organic compounds (VOCs). The challenge of our experimental setting is to deal with the problems of low selectivity, especially in normal operating conditions, so that ambiguous sensor responses (i.e. referable to more than one VOC) can be given, at least, a qualitative interpretation. In order to carry out the signal disambiguation task, a computational technique employing simple classifying rules and fuzzy descriptions has been engineered. The basic idea is that, if the same gas is actually measured by two or more sensors, then the estimated concentrations will show a low variance, with an accuracy related to the number of concordant sensors. Experiments show that, despite the cheapness of the setup and the coarse-grained nature of the provided response, encouraging results can be obtained and prospective work can follow.

A Comparative Evaluation of Retrieval Methods for Duplicate Search in Image Database

Abstract Visual database systems are continuously enriched by original and processed images derived from the original using editing techniques. This last type of image (called duplicate) represents a relevant quote of unnecessary stored images and, moreover a clear case of image similarity. In this paper a comparative evaluation between image retrieval methods in the specific case of duplicate search is proposed. All methods are based on low-level features extraction, such as colors, shape and patterns, and are suitable in automatic systems. The performance comparison is made in terms of effectiveness using a database of 6368 natural pictures and 12 test sets of 18 images each.

Discriminating gaseous emission patterns in low-cost sensor setups

This work presents a two-step heuristic that employs extremely low-cost sensors for gaseous emission event discrimination. These events are triggered by particular patterns of sensor responses possibly occurring when a certain gas is emitted; patterns are then used to produce human-understandable inference rules describing the kind of emission measured. The technique, challenged by the high cross-sensitivity of the employed sensors, is based on two steps: first, sensor response patterns are extracted (unsupervisedly) from measurement signals by means of a recently proposed computational intelligence technique; second, a ‘credibility index’ is applied (supervisedly) to each pattern via fuzzy membership functions. The outcome is a set of IF THEN statements weighted by fuzzy constraints. Experiments show that such inferences allow for accurate gaseous emission event discrimination.

Dialogue-oriented interface for linguistic human-computer interaction: A chat-based application

This work presents a prototype conversational system enabling human-computer interaction by using natural language expression. As an enhancement to well-known conversational agents like chatbots, in the proposed setting, human-machine dialogue is intended as a query/answer monotonic process aimed at minimizing semantic ambiguity within communication and delivering the required service. When user queries are ambiguous, hence semantically distant from the set of possible recognized interpretations, the system instantiates a dialogue with the user. In this case, the system provides suggestions on how to reformulate the query until a valid form is reached; this feed-back makes the dialogue-oriented interaction process resemble an ordinary chat (in the very restricted domain of system services) but with a machine interlocutor. The popularity of the chat as a synchronous communication instrument lets our proposal be suitable for a great variety of applications.

An ontology-based approach to human telepresence

Detecting human presence automatically is a challenging task since several environmental parameters may affect the quality and the continuity of detection. Although many techniques have been developed so far in the literature to solve this problem, they generally rely on well-defined operational context. Hence, they are sensitive to uncontrolled variables and unpredicted events. In this work an ontology-based approach to human telepresence detection is presented. Contrarily to classic sensor-driven techniques, a top-down methodology is applied. Starting from a formal description of the problem ontology, a set of high-response rate and low-response rate sensors is employed in a computational model. As a consequence of this model, a multi-sensor equipped device has been experimentally setup to conduct measurements on real scenarios. Experiments have been devised to estimate the robustness of the detection. In particular, some preliminary evaluations related to using a minimal set of chemical sensors are reported.