BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sketch Recognition with Multiscale Stochastic Models of Temporal P
 atterns - Dr Metin Sezgin\, Computer Laboratory\, University of Cambridge
DTSTART:20061026T131500Z
DTEND:20061026T140000Z
UID:TALK5633@talks.cam.ac.uk
CONTACT:Daniel Bernhardt
DESCRIPTION:Sketching is a natural mode of interaction used in a variety o
 f settings. For example\, people sketch during early design and brainstorm
 ing sessions to guide the thought process\; when we communicate certain id
 eas\, we use sketching as an additional modality to convey ideas that can 
 not be put in words. The emergence of hardware such as PDAs and Tablet PCs
  has enabled capturing freehand sketches\, enabling the routine use of ske
 tching as an additional human-computer interaction modality. \n\nBut despi
 te the availability of pen based information capture hardware\, relatively
  little effort has been put into developing software capable of understand
 ing and reasoning about sketches. To date\, most approaches to sketch reco
 gnition have treated sketches as images (i.e.\, static finished products) 
 and have applied vision algorithms for recognition. However\, unlike image
 s\, sketches are produced incrementally and interactively\, one stroke at 
 a time and their processing should take advantage of this.\n\n\nIn this ta
 lk\, I will describe ways of doing sketch recognition by extracting as muc
 h information as possible from temporal patterns that appear during sketch
 ing. I will present a sketch recognition framework based on hierarchical s
 tatistical models of temporal patterns. I will show that in certain domain
 s\, stroke orderings used in the course of drawing individual objects cont
 ain temporal patterns that can aid recognition. Build on this work\, I ill
 ustrate how sketch recognition systems can use knowledge of both common st
 roke orderings and common object orderings. I will present a statistical f
 ramework based on Dynamic Bayesian Networks that can learn temporal models
  of object-level and stroke-level patterns for recognition. This framework
  supports multiobject strokes\, multi-stroke objects\, and allows interspe
 rsed drawing of objects – relaxing the assumption that objects are drawn
  one at a time. The system also supports real-valued feature representatio
 ns using a numerically stable recognition algorithm. I will present recogn
 ition results for hand-drawn electronic circuit diagrams. The results show
  that modeling temporal patterns at multiple scales provides a significant
  increase in correct recognition rates\, with no added computational penal
 ties.\n\n\n
LOCATION:Rainbow Room\, Computer Laboratory
END:VEVENT
END:VCALENDAR