Object removal in movie post-production

This page contains demonstration video clips showing the results of a new algorithm that allows interactive object removal. The user must delineate a matte around the object (the Rig) to be removed (a garbage matte) and the algorithm then employs a recursive Bayesian approach for reconstructing the motion underlying the Rig. Having reconstructed the motion, image information from future and past frames can then be used to reveal the picture hidden by the object in each frame. The only information provided by the user is the matte. The algorithm then builds frames automatically using a combination of standard motion estimation techniques and a Bayesian refinement strategy for motion interpolation.

The clips shown here are downsampled from 720x576 to 180x144 for ease of downloading. All images on this page copyright The Foundry (www.thefoundry.co.uk)

Figure 1: Removing a Motorcycle: Orignal Clip, User Matte, Result of Automated RigRemoval
Figure 2: Removing a Person: Orignal Clip, User Matte, Result of Automated RigRemoval

All images on this page copyright The Foundry (www.thefoundry.co.uk)

Bibliography

  • A Bayesian Framework for Recursive Object Removal in Movie Post-Production. A. Kokaram , B. Collis and S. Robinson (2003) In (ICIP'03). Barcelona Spain. ( pdf )
  • Filling in the gaps. B. Collis and A. Kokaram (2004) IEE Electronics Systems and Software. ( pdf )
  • Automated rig removal with Bayesian motion interpolation. A.C. Kokaram , B. Collis and S. Robinson (2005) Proceedings of the IEE Journal on Vision, Image and Signal Processing, 152:407-414. ( pdf )
Page last modified on December 17, 2007