Image and Video Matting (ICCV 2013 Tutorial)
Date:  2:00 PM - 6:00 PM, December 1, 2013
Location: International Conference on Computer Vision (ICCV 2013), Sydney, Australia.
Organizers:
Speakers:
Tutorial Description: 

Matting refers to the problem of accurate extraction of foreground objects in images and video, which plays a fundamental role in image and video editing operations. Using matting techniques to create novel composites or to facilitate other editing tasks such as photo enhancement, red-eye correction, foreground object retargeting, and non-photorealistic rendering, has increasingly garnered interest from both professionals and customers alike. In this tutorial, we detail the current state-of-the-art algorithms and trends in image and video matting, review the performance evaluation system, and discuss current challenges and future directions in matting.  The tutorial is targeted toward graduate students familiarizing themselves with the field.
Tutorial Schedule:
  1. Introduction to alpha matting (2:00 – 2:50) [slide]:
    An introduction to the image matting problem is presented. This talk states the basic alpha matting problem and covers a number of natural image matting algorithms, specifically closed form matting, and robust and improved color sampling methods. Furthermore, the performance evaluation system (alphamatting.com) is introduced. Emphasis is given to the acquisition of ground truth datasets and to perceptually motivated error metrics. 

  2. Image matting limitations and challenges (2:50 – 3:40) [slide]:
    The sampling-based image matting methods are reviewed in greater detail and their limitations are presented. 

  3. Video matting (4:10 – 4:40) [slide]:
    The differences between the image and video matting problems are presented and several recent video matting methods are discussed in detail. 

  4. Future challenges in matting (4:40 – 5:10) [slide]:
    The current shortcomings of image and video matting  are reviewed and potential directions/avenues for future work are outlined