In our time, images and movies for cinema and television are frequently altered during the post-production stage to replace and complete unwanted areas. For example, a hateful graffiti present in an image is deleted. To produce quality results, it is important that the completed region has a realistic appearance and shows signs of aging. State-of-the-art methods addressing this problem are not suitable since they require the artist to use complex parameters and usually handle only one aging effect. The completion can also be done on a video in which a sound boom was inadvertently filmed. The filling of missing region in a video poses additional challenges, enforcing spatio-temporal consistency and the sheer amount of information to name a few, and current approaches are inadequate. Indeed, most methods, including those based on random Markov fields, cannot directly handle high definition video within a reasonable time. Furthermore, currents methods are limited by the type of camera movement, the missing region size, and the variation of light intensity.
An objective of this thesis is to develop a framework that allows artists to generate new aging effects based on a sample image containing an example of the desired effect. To achieve this, a novel texture synthesis approach adapted to the aging context is introduced. This approach does not need the artist to specify any complex parameter. A second objective is the development of a system that completes the missing regions of high definition video. A texture synthesis algorithm using a hole-filling approach is adapted by taking advantage of a coherence principle and a local search. Furthermore, the last part of this thesis presents a video completion method based on scale-invariant features tracking. This method is able to complete very large missing regions from video footage with non-trivial camera movements.
The results obtained from the different contributions of this research project show increased realism during the completion of missing areas of images and video sequences. The different methods are easy to use and intuitive because no complex parameter needs to be specified by the artist. Moreover, they are suitable for the iterative process of the artist. Finally, these methods fit well in the studio production pipeline since they require small computation times.
| Date | 20 Apr 2017 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Eric Paquette (Supervisor) |
|---|
Benoit, J. (Author),
Paquette (Supervisor),
20 Apr 2017Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering