A Digital Film-Saver for Archives

by George L. Kovács and Iván Kas

Pressure is growing fast for film archives to restore and save their priceless collections before the original material has decayed beyond repair. In the framework of the Hungarian National Research and Development Programs, SZTAKI leads a project consortium that will develop a complete and affordable digital film-to-film restoration system for archives that is not currently available on the market.

According to UNESCO, there are 2.2 billion metres of 35mm movie film in archives all over the world. In spite of the huge efforts of special, air-conditioned storage spaces, most of these films will soon be completely useless due to their extreme sensitivity to various physical and chemical effects and their current level of degradation. The amount of film to be saved in Hungary is about 60 million metres. Ninety percent of these are sound films, two-thirds are positive, one-third are negative and about fifty percent were taken in Hungary. These films are part of the national heritage and as such must be restored and saved.

Similar efforts exist worldwide, with some results having been achieved in the USA, Japan and Europe, including Kodak's CINEON system and the FRAME system developed by a European project consortium. However, complete systems are not available on the market, and parts such as scanners and film-writers are extremely expensive.

In Hungary, limited results were achieved by physical and chemical cleaning, rewinding and making one-to-one film copies, but in general these procedures finally 'kill' the 'wounded' originals. The reasons for commencing such an R&D effort in Hungary included the fact that earlier we did not have the appropriate mathematical/software knowledge for complete restoration, and the necessary high speed/high memory capacity computational tools have become available only recently. Furthermore, the success of the project can be guaranteed by our strong, well-balanced consortium (Hungarian National Film Archives, our academic research institute, the Veszprém University and the Cortex Ltd. - an SME), which has experience and knowledge in films, project management, laser technique, hardware and software making and in picture-, sound- and colour-processing.

Goals, Expectations and Preliminary Results
The project started in mid-2001 and will last for three years. Its goal is to produce an intelligent and powerful workstation with dedicated peripheries to restore and save - in real time - archive films damaged in their physical and/or informational condition. We began to create programs with mathematical methods - image- and sound-processing algorithms that will also be useful in other fields (eg medicine, biology, chemistry etc).

Movie picture experts do not accept electronic storage such as video, CD, DVD, HDVD etc, as final media for saved or restored films. As such, our solution should be based on film material. One of the reasons for this is that electronic media may have a limited lifetime, and it is obviously desirable to avoid compatibility problems wherever possible. As intermediate and/or mass production results however, all media will be used in our work.

To demonstrate the difficulties inherent in this project, we list some problems and the planned solutions.
One of these is the issue of film resolution. Recent TV screen resolution (about 1K) is insufficient, and while 2K or 4K are accepted for electronic storage as there is no better solution, filmmakers would prefer a resolution of 16K, which is not yet a realistic goal. Our basic aim is 6K resolution, as we are convinced that this is good enough to preserve all the picture/colour parameters of an original film. To provide this resolution, the storage of one hour of film material would require between 10.5 and 74.5 TBytes capacity (see Table). It is also evident that in the case of a resolution of 6K with the given colour densities, one metre of film needs 6.4 Gbytes minimal background memory, which leads to approximately 2 Tbytes for 300 metres. The minimal background memory capacity is defined by the fact that film professionals do not accept anything less than 300m of film as a basic unit. This value - 2 Tbytes - is still a minimal value, and is more or less the maximum available today.

Table: Some resolution/mass storage capacity request data.

Another critical issue is the speed of the film restoration process. The processing time of a frame means the input, output and total processing time of a frame, plus the time required for data transfers within the system. Our plan is to bring this down to 60 seconds or less. The best result so far is 72 seconds, achieved with the FRAME system developed by a European project consortium led by Joanneum Research.

Using this speed, one system is able to restore ten movies per year. We expect to compensate for missing or damaged frames and to correct scratches and patches on the films. We will be able to compensate for shakes or vibrations due to camera movements, which affect the viewing quality and the value of the film. We expect to produce a film that is almost as good as the original could have been in every sense, including richness of detail, colour, contrast etc. We plan to correct and compensate for all dirt, patches and perforations due to bad handling and chemical erosion of the film. Earlier, these could only be partially corrected by dangerous chemical processes, which have several negative side effects. Emulsion problems, copying errors, wet materials effects etc, will also be corrected and eliminated with the digital processing.

Figure 1: A typical worn-out film-frame with damaged perforation.	Figure 2: The experimental set-up of the film-saver with scanner and sound correction (in the middle) and film-writer (completely covered on the left-hand side).

The basic architecture of the system is rather simple. An input unit (scanner: high resolution line-camera) reads the film to be improved, a processing unit (high-capacity workstation with appropriate software to meet the above expectations) performs the restoration job and an output unit (film-writer based on a high-speed rotating mirror with modulated RGB laser beams) produces the new film. A terabyte-sized background memory (discs) stores data and is used as the basis for electronic read/write procedures, and a high-speed data network makes communication possible. Since we deal with sound restoration as well, sound management is symmetrical to picture management, but it works completely differently.

Figure 3: Experimental result with colour and contour corrections.

Since the project has been running for over a year, several results have already been achieved which prove that we are on the right track. All basic error types can be managed with the experimental software packages, and the software system plan is ready. All basic hardware units have been designed, several experimental parts are working already, and other parts are currently being constructed.
The consortium is open for a broader cooperation in the future.

Please contact:
George L. Kovács, SZTAKI
Tel: +36 1 279 6140
E-mail: gyorgy.kovacs@sztaki.hu

Iván Kas, SZTAKI
Tel: +36 1 279 6270
E-mail: kas@luna.ikk.sztaki.hu