|go to week of Sep 29, 2013||29||30||1||2||3||4||5|
|go to week of Oct 6, 2013||6||7||8||9||10||11||12|
|go to week of Oct 13, 2013||13||14||15||16||17||18||19|
|go to week of Oct 20, 2013||20||21||22||23||24||25||26|
|go to week of Oct 27, 2013||27||28||29||30||31||1||2|
Decision and Control Lecture Series
Decision and Control Laboratory, Coordinated Science Laboratory
Controlled Active Vision/Image Processing with Applications to
Medical Image Computing
Professor, Computer Science and Applied Mathematics/Statistics
SUNY, Stony Brook University
Wednesday, November 13, 2013
3:00 p.m. to 4:00 p.m.
In this talk, we will describe some theory and practice of controlled active vision The applications range from visual tracking (e.g., laser tracking in turbulence, flying in formation of UAVs, etc.), nanoparticle flow control, and sedation control in the intensive care unit. Our emphasis will be on the medical side, especially image guided therapy and surgery. This includes projects such as radiation planning in cancer therapy, traumatic brain injury, and left atrial fibrillation. Accordingly, we will describe several models of active contours for which both local (edge-based) and global (statistics-based) information may beincluded for various segmentation tasks. We will indicate how statistical estimation and prediction ideas (e.g., particle filtering) may be naturally combined with this methodology. A novel model of directional active contour models based on the Finsler metric will be considered that may be employed for white matter brain tractography. Very importantly, we will describe some ideas from feedback control that may be used to close the loop around and robustify the typical open-loop segmentation algorithms in computer vision.
In addition to segmentation, the second key component of many active vision tasks is registration. The registration problem (especially in the elastic case) is still one of the great challenges in vision and medical image processing. Registration is the process of establishing a common geometric reference frame between two or more data sets obtained by possibly different imaging modalities. Registration has a substantial literature devoted to it, with numerous approaches ranging from optical flow to computational fluid dynamics. For this purpose, we propose using ideas from optimal mass transport (Monge-Kantorovich). The optimal mass transport approach has strong connections to optimal control, and can be the basis for a geometric observer theory for tracking in which shape information is explicitly taken into account. Finally, we will describe how mass transport ideas may be utilized to generate hexahedral meshes with applications to problems in biomechanics.
The talk is designed to be accessible to a general applied mathematical/engineering audience with an interest in vision, control, and image processing. We will demonstrate our techniques on a wide variety of data sets from various medical imaging modalitie
SEMINAR RECEPTION IN NCSA LOBBY IMMEDIATELY FOLLOWING