Project: Automatic Image Restoration

Film is a fragile medium, prone to damage and decay; even newly shot film can suffer from scratches or other damage while it is being processed. There is a vast amount of film in archives, which cannot be exploited in its present form because it is too expensive to repair and restore, either physically or digitally. Nearly all rights holders would like to exploit their back catalogue commercially, but the cost of restoration methods is often greater than any likely commercial return._x000D_AIR will research and develop new technologies, automated processes and systems to restore digital representations of damaged film, COly from archives but also for postproduction. The goal is to develop tools that can automatically process the digital files from scanned film in real time (24 frames per second) or even faster, with little or no human intervention, thereby radically reforming the economics of film restoration. _x000D_Film is the oldest method of recording moving images. Almost everybody has an emotional engagement with ‘the movies’ - they entertain, educate, and amuse us. We laugh, cry, or gasp at their contents. Hollywood alone has produced over 300,000 movies in the past 100 years. The vast majority of film shot still exists in some state or other - but little of it is useful, due to damage and decay. The catalogue of film defects includes problems with film grain, dirt, negative sparkles, gelatin loss, emulsion loss, base scratches, emulsion scratches, film hair, film scuffing, colour fading, flickering, unsteadiness, and mould. The current approach to restoration is to clean and mend the physical film so that it can be scanned digitally, and then restore the digital image file using a wide range of sophisticated image processing methods. The resulting restored digital file can then be used to print new viewing copies, or for digital projection or distribution on the Internet or disc. Unfortunately, digital restoration is still a complex and time consuming process, which needs skilled operators to control each stage._x000D_AIR proposes a radical new approach to automating and speeding up the processes of analysing and restoring the digitised film. The process will involve the steps of: _x000D_- Evaluating the defects of the material that are apparent in the digital file and automatically applying corrections. _x000D_- Making a second analysis of the repaired file, to assess the further corrections that need to be applied to colour and tone grading_x000D_- Automatically re-grading and colour correcting the digital film_x000D_- Outputting the new corrected and optimally graded movie, in whatever format is required._x000D_Automating the analysis, defect correction and grading processes will involve systematically codifying the decisions and corrections currently applied by skilled restorers, to generate a library of control signals for automatic application. Automatic re-grading will require the application of novel algorithmic methods based on colour constancy and Retinex Theory. ‘Real time’ operation of the process will require implementation in a PC architecture with FPGA and GPU hardware acceleration._x000D_The outcome of the project will be a combined turnkey hardware and software system, which will accept and work with the digital output from any film scanner or Telecine machine. The customers for the system will be film and television archives (which hold large amounts of film), postproduction companies and digital film laboratories._x000D_Restored films will find market outlets in the ever expanding number of broadcast channels (the European Audiovisual Observatory reported that Europe already had 6500 channels by 2008), on-line entertainment, for cable, satellite, and download, IPTV, DVDs, cinema viewing, and subscription services such as ‘Lovefilm’ or ‘Netflix.’ _x000D_The Pship comprises two research performing SMEs (Cintel and Marquise) and the world's oldest film archive (Cinecitta LUCE). Cintel and Marquise are acknowledged experts in film scanning and digital processing. Cintel has over 80 years of experience in film related products and sells a suite of manually controlled tools for digital film restoration, which run of an FPGA-accelerated architecture. Marquise makes a software colour corrector, for grading images under operator control. It works closely with researchers at EPFL (which will be a subcontractor) who are world-leading experts in colour perception and automated colour processing. Cinecitta LUCE provides immense expertise and experience in the problems and processes of film restoration._x000D_

Acronym AIR (Reference Number: 6439)
Duration 01/09/2011 - 31/08/2013
Project Topic AIR will research and develop new technologies, automated processes and systems to restore digital representations of damaged film, COly from archives but also for postproduction, to facilitate its economic exploitation in the cinemas, on TV, over the internet or on disc.
Network Eurostars
Call Eurostars Cut-Off 6

Project partner

Number Name Role Country
3 Cinecitta Luce S.p.A Partner Italy
3 Cintel International Ltd Coordinator United Kingdom
3 Marquise Technologies sarl Partner Switzerland