Ikerbasque researcher: Fadi Dornaika
Recogniton of subtle facial expression in photographs and videos
Your have a patent in Attentive panoramic sensing for visual telepresence. What does that stands for?
We have introduced a new attentive panoramic sensor conceptually based upon the human foveated visual system. The proposed sensor consists of a panoramic video sensor and a high-resolution camera mounted on a pan/tilt platform. This sensor can be a solution to the Field of View/Resolution tradeoff. We have proposed a framework for automatically combining high-resolution images (fovea) with low-resolution panoramas. The attentive sensing is thus obtained from fusing the high resolution small field of view image with low resolution panoramas. Such systems may find application in surveillance and telepresence systems that require a large field of view and high resolution at selected locations.
Now you are working on subtle facial expression recogniton in photographs and videos. What techniques are used to get those subtle expressions in computers?
Emotion is fundamental to human experience, influencing cognition, perception, and everyday tasks. Affective computing attempts to bridge the gap that typical human-computer interaction largely ignored. In the vision community, still images and videos depicting faces constitute the main channel that conveys human emotion. Facial expressions are generated by contractions of facial muscles, which result in the deformation of facial features such as the eyelids, eyebrows, nose and lips. Early automatic facial expression recognition systems used simple geometric models, but the recognition process has now matured into a science of sophisticated mathematical tools and matching processes.
Subtle facial expressions are shown when the muscles are not contracted very much and are at a lower intensity. Building automatic facial expression recognition systems faces two major challenges. The first challenge is to extract from the raw visual data (images and videos) features that are good indicators about the displayed facial expression. The second challenge is to propose a scheme capable of distinguishing between several expressions. To address these challenges a plethora of efforts have been made. Statistics and machine learning algorithms have been used for modeling the complex human facial expressions.
The facial action coding system (FACS) provides the most widely used method to measure facial movement. In the FACS, a face is divided into 44 action units (AUs) according to their locations as well as their intensities. A combination of AUs is used to model the universal expressions. Recent works adopted the extraction of 3D facial dynamics in order to make the expression recognition feasible even in the presence of head motions. Automatic facial expressions in video sequences can use shape deformations, texture dynamics or a combination of them.
What has been the main difference between working on the French National Geographical Institute and now being a profesor at the University of the Basque Country?
The main difference between the French National Geographical Institute (IGN) and the University of the Basque Country (UPV) is the type of research performed. IGN is strongly industry-related, which implies that projects have a fixed topic. For this reason, from IGN point of view, the research has a strong applied component, being oriented more towards development rather than exploring novel algorithms and ideas. Personally, due to my professional experience and career, I prefer more the academic research, when one has the freedom to pursue his own research interests.
Share this:
