Semantic Media Retrieval
Semantic Media Retrieval is about the search of knowledge from multimedia data sources. In our research group we focus the attention to images and videos.
People describe their life experience using words, that is a representation that could have many differences, when we consider different persons, due to the users’ knowledge and experience and because of the context. This represent a “semantic gap” between the conceptualizations of the world expressed using language, and the experience of the world, whose most direct representations are photos and media in general. Due to this, current media understanding systems are still very much example-driven (e.g., find photos similar to a given one on the basis of a set of features).
Content and Context Based Indexing
Event Based Image and Video Understanding
Gamification of Media Retrieval
The pervasive availability of the Internet, allied with the spread of increasingly powerful digital facilities, has led digital multimedia to be the primary source of visual information in many aspect of our society, including media, politics, national security and advertisement. However, the historical assumption that photographs can be trusted as a true representation of reality does not hold anymore. Nowadays, affordable and sophisticated graphics editing software allow for the creation of sophisticated and visually compelling photographic fakes, which easily puzzle our perception of reality. Trustworthiness of the information conveyed by digital media is becoming one of the key challenges for our information society, strongly affecting the success and penetration of future multimedia applications. The urgent need of efficient techniques to cope with security issues related with multimedia data motivates the MMLab research. Multimedia forensics techniques are particularly relevant as they deal with the recovery of information that can be directly used to authenticate and estimate the trustworthiness of digital multimedia contents.
Computer Vision and Behavior Analysis
Computer vision aims at providing computers the ability to understand the content of images and videos on a high level. In an ideal system, computer vision allows machines to understand and automate tasks that the human visual system can do.
Nowadays feature-based methods are broadly used along with machine learning techniques and optimization frameworks. In recent years, Deep Learning techniques have taken the computer vision field by storm. The accuracy of deep learning algorithms on several benchmark tasks such as classification, segmentation, and optical flow has surpassed prior methods.
Human Behaviour Analysis (HBA) is more and more being of interest in Computer Vision and Artificial Intelligence researchers. Among other applications, the MMlab focuses on human-centered applications, like Crowd Analysis and Ambient–Assisted Living (AAL), which have been in great demand in recent years.
Crowd Analysis and Simulation
Trajectory Prediction in Crowded Scenarios
Human Pose Estimation
Automatic Highlights Generation