1 / 36

Object Vision 2

Object Vision 2. PSY 295 – Sensation & Perception Christopher DiMattina , PhD. Object recognition. A world of identifiable objects. Inferotemporal cortex. Neurons selective for very complex stimuli like faces. How do we get such fancy neurons?. Visual processing hierarchy

vinnie
Download Presentation

Object Vision 2

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Object Vision 2 PSY 295 – Sensation & Perception Christopher DiMattina, PhD

  2. Object recognition PSY 295 - Grinnell College - Fall 2012

  3. A world of identifiable objects PSY 295 - Grinnell College - Fall 2012

  4. Inferotemporal cortex • Neurons selective for very complex stimuli like faces PSY 295 - Grinnell College - Fall 2012

  5. How do we get such fancy neurons? • Visual processing hierarchy • Mid-level vision asks how we go from simple edge detectors to neurons sensitive to complex objects • Problem of region labeling and grouping PSY 295 - Grinnell College - Fall 2012

  6. Pandemonium model PSY 295 - Grinnell College - Fall 2012

  7. Web activity: Pandemonium • http://sites.sinauer.com/wolfe3e/chap4/pandemoniumF.htm PSY 295 - Grinnell College - Fall 2012

  8. Hierarchical processing • Hubel and Wiesel’s investigations of primary visual cortex suggest hierarchical processing models • Center surround cells combine to form simple cells • Simple cells combine to form complex cells • Increasing feature selectivity, less position-dependence PSY 295 - Grinnell College - Fall 2012

  9. Ventral stream model PSY 295 - Grinnell College - Fall 2012

  10. Middle vision and object recognition PSY 295 - Grinnell College - Fall 2012

  11. Perceptual committees • Mid-level vision is optimized to detect specific features – committee of experts • Committee outputs are combined according to rules to arrive at decision PSY 295 - Grinnell College - Fall 2012

  12. Committee rules • Perceptual committees almost always agree on a single unambiguous interpretation • Exceptions that prove the rule PSY 295 - Grinnell College - Fall 2012

  13. The accidental viewpoint • Our perceptual committees avoid interpreting the scene as viewed from one exact location PSY 295 - Grinnell College - Fall 2012

  14. Accidental viewpoint demo • http://sites.sinauer.com/wolfe3e/chap4/ambiguityF.htm PSY 295 - Grinnell College - Fall 2012

  15. Edges belong to objects • World is comprised of objects which `own’ edges • Which object do we assign edge to? • Figure-ground assignment PSY 295 - Grinnell College - Fall 2012

  16. Border ownership cells in V2 PSY 295 - Grinnell College - Fall 2012

  17. Principles for figure-ground assignment • Surrounded-ness • Size • Symmetry • Parallelism PSY 295 - Grinnell College - Fall 2012

  18. Extremal edges PSY 295 - Grinnell College - Fall 2012

  19. Occlusion complicates things PSY 295 - Grinnell College - Fall 2012

  20. Occlusions • When is something occluded interpreted as a single object? • Concept of relatability PSY 295 - Grinnell College - Fall 2012

  21. T-junctions signal occlusions PSY 295 - Grinnell College - Fall 2012

  22. Mid-level vision • Use prior knowledge about statistical regularities in the natural environment to work backwards from 2-D image to correct 3-D object interpretation PSY 295 - Grinnell College - Fall 2012

  23. Object recognition & neural codes PSY 295 - Grinnell College - Fall 2012

  24. Template matching PSY 295 - Grinnell College - Fall 2012

  25. Impractical – need a lot of templates! PSY 295 - Grinnell College - Fall 2012

  26. Alphabet of complex features PSY 295 - Grinnell College - Fall 2012

  27. 2D shape cells in V4 PSY 295 - Grinnell College - Fall 2012

  28. 3D shape tuning in IT PSY 295 - Grinnell College - Fall 2012

  29. Human cortical specialization PSY 295 - Grinnell College - Fall 2012

  30. Structural Description PSY 295 - Grinnell College - Fall 2012

  31. Structural description • One way to get around problems with template matching is to use the fact that objects share a common structure • Match image to the structural description, i.e. specify in terms of parts and relationships. • Biederman’s “recognition-by-components” theory PSY 295 - Grinnell College - Fall 2012

  32. Geons • Alphabet of geometric primitives from which objects build • Structural description theory suggests object recognition should be view-point invariant PSY 295 - Grinnell College - Fall 2012

  33. Viewpoint dependence • Experiments show object recognition is viewpoint dependent • Faster for familiar viewpoints PSY 295 - Grinnell College - Fall 2012

  34. Monkey experiments PSY 295 - Grinnell College - Fall 2012

  35. Top-down influences • Debate whether object recognition feed-forward or top down • Object substitution masking is when perception of a briefly flashed object is blocked by a subsequently presented object • Suggests top-down re-entrant processing PSY 295 - Grinnell College - Fall 2012

  36. Web activity • http://sites.sinauer.com/wolfe3e/chap4/objectsubF.htm PSY 295 - Grinnell College - Fall 2012

More Related