• Top page
  • Timetable
  • Per session
  • Per presentation
  • How to
  • Meeting Planner



Neural Network Modeling

開催日 2014/9/13
時間 18:10 - 19:10
会場 Room G(303)
Chairperson(s) 酒井 裕 / Yutaka Sakai (玉川大学脳科学研究所 / Brain Science Institute, Tamagawa University, Japan)
濱口 航介 / Kosuke Hamaguchi (京都大学大学院医学研究科生体情報科学講座 / Department of Biological Sciences, Graduate School of Medicine, Kyoto University, Japan)

3D shape estimation from a single glossy object image

  • O3-G-2-4
  • 下川 丈明 / Takeaki Shimokawa:1 西尾 亜希子 / Akiko Nishio:2 佐藤 雅昭 / Masaaki Sato:1 川人 光男 / Mitsuo Kawato:1 小松 英彦 / Hidehiko Komatsu:2,3 
  • 1:ATR脳情報通信総合研 / ATR-BICR, Kyoto, Japan 2:生理研 / National Institute for Physiological Sciences, Okazaki, Japan 3:総研大院・生命科学・生理 / The Graduate University for Advanced Studies (SOKENDAI), Okazaki, Japan 

Gloss and shading enhance our 3D shape perception. Therefore, the appearance of specular reflection (gloss) and the diffuse reflection (shading) is thought to be one of clues for 3D shape perception. While the neural processing of these clues has not yet been extensively studied, one previous study isolated the following relevant image information for 3D perception [1]. In the light reflection process, the surrounding illumination is compressed depending on the second derivative (i.e., surface curvature) of the 3D shape, and this compression direction can be extracted from the image by populations of simple oriented filters ("orientation field"). Consequently, the "orientation field" has rich information about the 2nd derivative of the 3D shape. It is likely that our visual cortex use this kind of information for 3D perception because similar orientation extraction is performed in the primary visual cortex.
Here, we have made the estimation algorithm utilizing the "orientation field" as the 2nd derivative constraints to recover the 3D shape. We will show that 3D shape of a glossy object could be recovered by our estimation algorithm from a single and static image.

[1] R. W. Fleming, A. Torralba, and E. H. Adelson, Journal of Vision 4 (2004) 798.

Copyright © Neuroscience2014. All Right Reserved.