Inpainting also known as image reconstruction or image completion is a technique of correcting missing parts of the image in such a way that the final image appears more pleasing and visually plausible to an average observer.
Numerous methods have been developed since the year 2000 when prof. Marcelo Bertalmio introduced a term “Digital inpainting”. Previously inpainting was formulated as a disocclusion problem.
We research different inpating methods with a goal of using them to reconstruct old black and white photographs and films.
During the past decade the incredible technological progress made it possible for digital sensors (e.g. CCD, CMOS) to replace photographic film in most consumer and professional cameras. It is also a matter of just a few years before almost all cinema-quality films will be captured directly in the digital format without the need of physical light sensitive film.
There are however numerous archives around the world that hold more than a million feature films and television programmes all captured and stored on the physical film. In the era of youtube the main problem is not how to store those films in the digital format, but rather how to scan them so their quality is good enough for an average person to find interest in watching them. Once digitized they could reveal important historical information to a much wider audience, as rolls of films in the archives are essentially worthless, unless someone watches them.
If they are not digitized soon they may be lost forever due to progressing degradation. Many films made before 1951 have been lost in accidental fires because they were based on the Cellulose Nitrate – a very flammable material. These type of films are difficult and expensive to store, typically in room size refrigerators, so they should be digitized first.
The problem with digitizing any old film is simply that it is damaged to some extent. We concentrate on the two most common film ‘defects’ which are dust and scratches. They are present in any film, but their extend depends strongly on the past usage, handling and the film age.
This restoration of film from the defects should be done as the first step in film processing shortly after scanning. The motivation for reconstruction of these blemishes is twofold. First, they are simply disturbing to watch. Second, reconstructed film can be more easily treated with different filters used to reduce noise, film grain, frame registration, intra frame flickering and color degradation. These processes are necessary to allow better viewing experience.
The following classes of inpainting methods with be included in the comparison:
1. Extension of Texture Synthesis methods
2. Based on Partial Differential Equations (PDEs)
3. Patch-based (Exemplar-based)
4. Based on the theory of Compressive Sensing