Your face will be indexed. Be sure. Does it make you shudder? Ah-ah-ah. It is not sci-fi. Something so extended as Picasa is already capable to search and sort photos on face value. Yes, it is a great chance to populate ontologies but it gives the machines more power to know everything about you.
Face-finding has made huge improvements recent years. Researchers at Carnegie Mellon University have developed a system that can recognize faces in low resolution videos.
Freedom, privacy, information access… just think that when you upload something to the web you are providing more information than you thought.
The UNL, Universal Networking Language, is an ambitious initiative from United Nations that began in 1996. The UNL is an artificial computer language that replicates the functions of natural languages. It is one of the attempts to build a pivot computer language for machine translation.
Researchers from the group of Validation and Business Applications , based at Universidad Politécnica de Madrid’s School of Computing (FIUPM) have take advantage of UNL and its idea of Universal Word (UW) stems to build multilingual ontologies.
The use of universal words can reduce the ambiguity and diversity of ontologies, which are failing at providing universal representations of concepts within a domain because of its English centered nature.
It is worth a look, and it is quite cheap, just 25$. The next release or version is planned for 2010. It will necessary have to update all knowledge about Geospatial Data, Standards and Metadata that have emerged in the last years. Also the new challenges arised from Location Based Services and Mobile Geoweb have to be taken into account.
We have developed a new Geospatial ontology within the CINeSPACE project. The main idea behind this ontology is a more ample model of the spatial dimension. Some concepts, such as country, mountain or building are inherently spatial. But there exists an important number of concepts that are not physically geographic, although they are commonly related to certain areas. Some examples could be human activities as going shopping. Also events and parties or famous personalities are often related to a spatial extent. We have introduced the term “geoconcept” to name any entity with an inherently or indirectly associated spatial dimension. Some characteristics of the ontology are:
Its expressivity is OWL Lite. Fuzzy logic extensions were planned for spatial relationships, so simplicity is important to guarantee the decidability.
A rough set of spatial relationships is defined. As the spatial extent of abstract concepts is vague or uncertain, the RCC8 model can’t help with this modeling. Mereologic and neighbourhood properties are defined.
Spatial common reasoning is strongly related to sight sense. Many times we reference our position with what we see, although we don’t be exactly in that place. This effect is clearly seen in Geotagging processes, when sometimes the position of the photo’s target is provided, but other times the point where the photo was taken is recorded. We call this the source-target problem and we have added a new spatial relationship, hasNiceViewsOf, to model this problem.
Polygons centroid is added to the geometry part.
While complete fuzzy reasoners are developed, a simple model is provided to deal with different degrees of uncertainty, adding several fuzzy subproperties to spatial relationships. A set of SWRL rules is optionally added to manage fuzzy transitivity.
An skeleton of the ontology is depicted in the following figure:
CINeSPACE, experiencing urban film and cultural heritage while on the move, is a 6th Framework Programme european project in its second year. One of its main objetives is providing a rich media delivery platform for distributing specific, location-based urban, cultural and film information. The access to the content is semantically enhanced, so information can be filtered according to the context or user profiles.
As part of this project we are studing how humans interact with our physical context. The scenario changes when humans interact with the context, and with the multimedia available in that context. To model his scenario we should consider not only where we are but also where we are looking at.
We have analyzed geotagging process in social information spaces and found that there exists an ambivalent meaning for geographic points in such process. The geographic lat/long usually refers to the target of the video or photo, but sometimes the place where the photo was taken from is recorded. We call this issue the source-target problem in geotagging processes. Again, location and sight concepts are merged.
We have devised Geoconcepts ontology, which attempts at exploding the source-target problem. The results of this research will be presented at ICGIS 2008 conference.
Several new proposals are focusing on the “vision aware” context, and thanks to image recognition, providing related information. Some days ago Mans Shapshak introduced us Photosynth, a virtual 3D photo browser. It’s a pretty cool tool, which allows walking or flying through a virtual scene to see photos from any angle. But one thing is getting this working in a PC, and a different one to incorporate it to a mobile device.
Eyephone, in the other side, makes a realistic proposal for mobile device leaving all processing costs to server side. A demo is showed below. Using advanced object recognition from SuperWise Technologies they take as an input the GPS position, a photograph taken by the user, and the relative angle of vision to provide information about the location the user is looking at. You will think, It is perfect! Well, it would, if they had a prototype in the market…
We have made a small step in order to integrate vision and location to geospatial semantics, but more research is needed. Both meanings should be taken into account in systems which provide multimedia content while on the move, as in CINeSPACE.
but it gives the machines more power to know everything about you.
