Luca Canini

Emotional identity of movies

In the field of multimedia analysis, attempts that lead to an emotional characterization of content have been proposed. In this work we aim at defining the emotional identity of a feature movie by positioning it into an emotional space, as if it was a piece of art. The multimedia content is mapped into a trajectory whose coordinates are connected to filming and cinematographic techniques used by directors to convey emotions. The trajectory evolution over time provides a strong characterization of the movie, by locating different movies into different regions of the emotional space. The ability of this tool in characterizing content has been tested by retrieving emotionally similar movies from a large database, using IMDb genre classification for the evaluation of results.

Interactive storytelling via video content recombination

In the paper we present a prototype of video-based storytelling that is able to generate multiple story variants from a baseline video. The video content for the system is generated by an adaptation of forefront video summarisation techniques that decompose the video into a number of Logical Story Units (LSU) representing sequences of contiguous and interconnected shots sharing a common semantic thread. Alternative storylines are generated using AI Planning techniques and these are used to direct the combination of elementary LSU for output. We report early results from experiments with the prototype in which the reordering of video shots on the basis of their high-level semantics produces trailers giving the illusion of different storylines.

Markers of unsustainable gambling for early detection of at-risk online gamblers

In this work we propose novel markers for identifying at-risk gamblers based on the concept of sustainability. The first hypothesis here verified is that problematic gamblers oscillate between intervals of increasing wager size followed by rapid drops, probably because they exceed their economic sustainability limits. Due to the non-periodic nature of these fluctuations, the proposed marker detects a certain occurring feature, such as a rapid drop in wager size, over a wide range of fluctuation periods, drop sizes and shapes. The second marker, counting the number of games the gambler is involved in, aims at predicting possible consequences of an exceeding amount of time dedicated to gambling, that ultimately causes social and relational breakdowns. In the experimental phase we demonstrate how the adoption of these markers allows for identifying larger segments of high- and medium-risk gamblers with respect to previous research on actual betting behaviours.

Users' response to affective film content: A narrative perspective

In this paper, we take a human-centred view to the definition of the affective content of films. We investigate the relationship between users physiological response and multimedia features extracted from the movies, from the perspective of narrative evolution rather than by measuring average values. We found a certain dynamic correlation between arousal, derived from measures of Galvanic Skin Resistance during film viewing, and specific multimedia features in both sound and video domains. Dynamic physiological measurements were also consistent with post-experiment self-assessment by the subjects. These findings suggest that narrative aspects (including staging) are central to the understanding of video affective content, and that direct mapping of video features to emotional models taken from psychology may not capture these phenomena in a straightforward manner.

Estimating cinematographic scene depth in movie shots

In film-making, the distance from the camera to the subject greatly affects the narrative power of a shot. By the alternate use of Long shots, Medium and Close-ups the director is able to provide emphasis on key passages of the filmed scene. In this work we investigate four different inherent characteristics of single shots which contain indirect information about scene depth, without the need to recover the 3D structure of the scene as a prior step. Specifically, 2D scene geometric composition, frame colour properties, shot motion distribution and content are considered for classifying shots into three main categories. In the experimental phase, using SVM classifiers, we first test the ability of each single feature in distinguishing shot types; then we combine the whole feature set in order to improve the classification performance.

Classifying cinematographic shot types

In film-making, the distance from the camera to the subject greatly affects the narrative power of a shot. By the alternate use of Long shots, Medium and Close-ups the director is able to provide emphasis on key passages of the filmed scene. In this work we investigate five different inherent characteristics of single shots which contain indirect information about camera distance, without the need to recover the 3D structure of the scene. Specifically, 2D scene geometric composition, frame colour intensity properties, motion distribution, spectral amplitude and shot content are considered for classifying shots into three main categories. In the experimental phase, we demonstrate the validity of the framework and effectiveness of the proposed descriptors by classifying a significant dataset of movie shots using C4.5 Decision Trees and Support Vector Machines. After comparing the performance of the statistical classifiers using the combined descriptor set, we test the ability of each single feature in distinguishing shot types.

Multimodal Space for Rushes Representation and Retrieval

In video content analysis, growing research effort aims at characterising a specific type of unedited content, called rushes. This raw material, used by broadcasters and film studios for editing video programmes, usually lies un-annotated in a huge database. In this work we aim at retrieving a desired type of rush by representing the whole database content in a multimodal space. Each rush content is mapped into a trajectory whose coordinates are connected to multimodal features and filming techniques used by cameramen while shooting. The trajectory evolution over time provides a strong characterisation of the video, so that different types of rushes are located into different regions of the multimodal space. The ability of the proposed method has been tested by retrieving similar rushes from a large database provided by EiTB, the Basque Country main broadcaster.

A free Web API for single and multi-document summarization

In this work we present a free Web API for single and multi-text summarization. The summarization algorithm follows an extractive approach, thus selecting the most relevant sentences from a single document or a document set. It integrates in a novel pipeline different text analysis techniques - ranging from keyword and entity extraction, to topic modelling and sentence clustering - and gives SoA competitive results. The application, written in Python, supports as input both plain texts and Web URLs. The API is publicly accessible for free using the specific conference token1 as described in the reference page2. The browser-based demo version, for summarization of single documents only, is publicly accessible at http://yonderlabs.com/demo.