--------------------------------------------------------------------------------------------------------------------

                             My Research At Stanford University

                                        Salih Burak Göktürk
                                         Stanford University

--------------------------------------------------------------------------------------------------------------------

In this talk, I will present some highlights of the research that I 
conducted during my PhD at Stanford University. The main theme of my 
thesis is the recognition process. Recognition is one of the main tasks 
of both human and computer vision. We address the recognition process 
by a combination of two tasks: feature estimation and statistical 
classification. My projects fit in this framework, while we use different 
features for various applications.

The first project that I will discuss is the face-tracking project. 
We designed a model-based monocular face tracking system that tracks the 
pose and deformations of face in 3D. The output of the tracking system is 
a shape vector, which is used as the feature for a facial expression 
recognition system. These features are fed to a support vector machine 
classifier for the classification of the facial expressions.

The second project of this talk is the recognition of 3D shapes from 
computed tomography (CT) with application to detection of various cancers. 
For the feature estimation stage of this project, we designed a new 3D 
shape representation, called Random Orthogonal Shape Sections (ROSS) 
descriptor. This descriptor utilizes a random set of triples of mutually 
orthogonal planar sections through the shape. The statistics obtained 
over a sufficiently large number of random triples provides invariant and 
diagnostic descriptors (representation) of the shape. The descriptors 
are used as features by a support vector machine classifier for the 
recognition of various cancerous tissues.

Information about these projects and various publications can be found at:

http://robotics.Stanford.edu/~gokturkb