Research

By , November 21, 2013

These are the research projects I have been involved in. I have had the chance to work with many wonderful people.

Ongoing projects

EAGLE – Exploiting semAntic and social knowledGe for visuaL rEcognition

EC Marie Curie Actions The Marie Curie IOF grant EAGLE started in October 2014. The aim of this three-years project is to effectively exploit semantic and social knowledge for visual recognition. Marie Curie individual fellowships are highly prestigious and competitive and are meant to support the best, most promising European researchers. Project page.

Advisors: A. Del Bimbo (U Florence), Li Fei-Fei (Stanford)
Funded by the European Commission under the excellence program MSCA

 

Completed projects (but not forgotten)

Tag refinement and localization in web videos

Our framework for tag suggestion and localization We present a novel data-driven approach for automatic video annotation that effectively increases the number of tags originally provided by users, and localizes them temporally, associating tags to shots ().

Collaborators: M. BertiniG. SerraA. Del Bimbo

Data-driven methods for social media analysis and annotation

Tag Refinement Example We are working at a framework for evaluating nearest-neighbor methods for tag refinement and automatic image annotation. We performed extensive and rigorous evaluation using standard large-scale datasets to show that the performance of these methods is comparable to that of more complex and computationally intensive approaches ().

Collaborators: T. Uricchio, M. BertiniA. Del Bimbo

A SIFT-based method for copy-move forgery detection

A SIFT-based Method for Image Forensics Understanding if a digital image is authentic or not, is a key purpose of image forensics. We proposed an effective approach, based on SIFT features, for copy-move detection and localization ().

Collaborators: I. AmeriniG. SerraR. CaldelliL. Del TongoA. Del Bimbo
Press coverage: “La Repubblica” and “Corriere della Sera” (Jan 2014)

A Context-Dependent Kernel for Logo Recognition

A Context-Dependent Kernel for Logo Recognition We contributed through this work to the design of a variational framework able to recognize multiple instances of logos in image archives. Logos are seen as constellations of local features and matched by minimizing an energy function which captures feature co-occurrence/geometry ().

Collaborators: H. Sahbi (Telecom ParisTech), G. SerraA. Del Bimbo

Human action categorization in unconstrained videos

Human action categorization We introduced a novel local descriptor based on both image gradient and optic flow to respectively model the appearance and motion of human actions. We used also radius-based clustering with soft assignment in order to create an effective codebook model ().

Collaborators: L. SeidenariG. SerraM. BertiniA. Del Bimbo

Automatic trademark detection and recognition in sports videos

formula1 We developed a semi-automatic system for detection and recognition of pre-defined brands and trademarks in broadcast TV. The number of appearances of a logo, its position, size and duration is recorded to derive indexes and statistics that can be used for marketing analysis ().

Collaborators: M. BertiniA. Del BimboA. D. BagdanovA. Jain
Partially funded by Sport System Europe s.r.l.

Comments are closed

Panorama Theme by Themocracy