Reading: "Real-Time User Guidance for Freehand Drawing"

y Computer Graphics paper

http://research.microsoft.com/pubs/168814/shadowdrawsiggraph11.pdf


・Contributions

  • An interactive drawing system that dynamically adapts to the user's drawing and provides real-time feedback.
  • New technique of partial spatial matching; it allows for multiple matching images based on different sub-regions of the image
  • Unique verification stage and methods for determining the blending weights
  • New approach to facilitate users to draw more realistically proportioned line drawings

・Strength

  • Automatically blend relevant images from a large database to construct the shadows
  • Dynamically adapts to the user's drawing in real-time and produces suggestions accordingly
  • While preserving the essence of drawing such as freedom and expressiveness
  • Subject's personal style of drawing is maintained regardless of the use of ShadowDraw.

・Weakness

  • As seen from the "5.1 User Studies" section, the effect of ShadowDraw is limited to those with "average" drawing skills. In order to be useful to the "poor" group, more sophisticated system is required to understand what they are drawing, even if "the aspcet ratios and basic shapes of thier drawings were far off from those of the objects they were intending to draw."

・Ideas for extension

  • I was thinking how this technique can be utilized with wearable devices such as Google Glass. For example, the user with his Glass draws on normal paper, and the camera on his Glass captures what the person is drawing. According to the current drawing, the system produces suggestions on the glass, just like ShadowDraw does on the tablet interface. This extension doesn't restrict where to draw. Need more sophisticated technique to detect objects.
  • Adjust what sort of style of drawing the user wants to achieve? (Currently, ShadowDraw is only helpful for realistic drawing.)

・Open Questions

  • How can we assess the "aesthetic improvement"
  • Adjust what sort


・Function
1. Build a database from a collection of 30,000 images collected from the Web. Convert each image to an edge drawing using the long edge detector technique developed by [Bhat et al. 2009] and store each image.
2. Analyze, code, and store overlaping windows in each edge image
3. Convert each window to edge descriptors and further code as sketches with distinct hash keys using min-hash [Chum et al. 2008]. (ここまでoff-line)
4. As the user draws, ShadowDraw analyzes the strokes using a similar encoding to dtermine hash keys for overlapping windows for fast matching with the database of images.
（3,4がよくわからないのであとで確認）




・感想
"Teddy [Igarashi et al. 1999]"って至る所で目にする。すげえ。