Analysis, Synthesis and Retrieval of 3D faces

 

Contact: Mariano Phielipp

 

Project Details:

 

The problem with the current approach of modeling the perception of faces is that use mathematical models of similarities that ignore how human perceived. This is a crucial problem nowadays when computers are becoming ubiquitous and more powerful every day. To use the computer as a better tool we need to program them in a way that they understand us. For successful collaboration between a human and a computer in a task that involved the retrieval of 3D faces, they need to ‘perceive’ the 3D face in the same manner that we human do. An application that needs this problem solved is when a witness need to describe the perpetrator of a crime. Nowadays when mugshot databases are consulted the query is based on mathematical model of similarity using a “Query by Example” technique. The other used method is the creation of an artificial face and then used that face using some mathematical model of similarity. Secondly, the traditional models for measuring similarity do no take in account the fact that not two people perceive faces in a similar way. Based on psychological studies the “contact” hypothesis imply that we recognize faces from people that we were more in “contact” more accurately. Consequently, we are more bias for the encoding of features that are most useful for distinguishing among own-race race. The model I intend to develop will try to capture the perception of faces considering this fact. Finally, our main form of communication is language. Consequently, for a better communication between the computer and the human seems natural to rely on words to characterize the 3D faces.

Voronoi diagram

The outcome of this reseach should provide us with algorithms that fullfill 3 pourposes. The first one is retrieval of a 3D face from a database of 3D faces with a set of labels that describe the face. The second one is the synthesis of a 3D face give a set of labes that describe a face. The final pourpouse is that given a face the algorithm automatically generate a description of the 3d face. The labels, description of the face, will the one collected form the game players.

 

More Information