DAVID: Recognition of Monuments in Images

In the past few years, several research works have addressed the problems posed by vision--assisted navigation systems. In a typical scenario, a tourist, while exploring an urban environment, is allowed to point a mobile camera to a monument, take a photograph, send it to a service provider, and eventually receive detailed information about the identity and relevance of the monument, and also links to related potentially relevant monuments, places and museums in the neighboring area. To accomplish this task, the system must represent objects of interest (e.g. monuments) with a certain degree of invariance to several geometric and photometric phenomena, so as to support recognition of the object in the photograph regardless of the viewpoint, illumination condition, background and scaling.

In addition, in photographs of monuments a significant portion of the scene usually correspond to the background: many interest points are associated with the background rather than with the monument. If two or more objects of interest, from some viewpoint, have similar - if not the same - background, discriminating images of different objects by comparing the image interest points becomes impossible unless some criterion is available to distinguish between salient interest points corresponding to the object and those corresponding to the background.
Current research activities focus on:

  • Defining a model to measure the saliency of interest points
  • Defining a model to estimate the stability of interest points with respect to the viewpoint

3D Face Recognition

Face recognition has been an active research area in the last years, with a major emphasis targeting detection and recognition of faces in still images and videos. More recently, the increasing availability of three-dimensional (3D) data, has paved the way to the use of 3D face models to improve the effectiveness of face recognition systems. In fact, solutions based on 3D face models feature less sensitivity - if not invariance - to lighting conditions and pose.

We have developed on an approach to 3D face description and matching based on the use of iso-geodesic stripes to capture distinguishing facial traits. Iso-geodesic stripes of the face are identified by measuring distances of surface points to a fiducial point located on the nose tip. Facial information captured by iso-geodesic stripes is then represented in a compact form by extracting the basic 3D shape of each stripe and evaluating the spatial relationships between every pairs of stripes. Mutual arrangements between pairs of iso-geodesic stripes are encoded in a graph structure, nodes corresponding to stripes and arcs between two nodes labeled with the description of the spatial arrangement between the corresponding stripes.
Current research activities focus on:

  • Investigation of new models to represent and compare facial stripes
  • Use of feature selection methods to investigate the relative relevance of stripe pairs, particularly in case of non-neutral facial expressions