Photographic film and its interaction with light : detection of dust and scratches for image restoration

A considerable portion of the cultural heritage of the last century is constituted by images impressed on photographic film, the majority of which are frames of motion picture films.
At present time, a large part of this heritage is at risk of being lost; this loss would break historical links, which are important for the economic, social and cultural development of future generations. The current ''digital revolution' has already produced drastic changes in image production technology, and 'classical photography' is disappearing; this leaves the field open for new powerful technical opportunities, but, at the same time, poses huge dilemmas for long-term archiving. The demise of classical photography is taking place in a precipitous and often ruinous way, causing several historied companies to go bankrupt.
Today's information society needs efficient and economic solutions for the digitization of this photographic heritage. The costs of the processes determine the amount of films that can be digitized, restored and made available to the community.
Research about digital movie restoration began around 1990, when it became possible to scan movie films in 2K resolution, thus providing high enough quality for the digitized images of a 35mm movie film. In recent years, several research projects about digital movie restoration have been undertaken. The research-project AURORA concentrated on the real-time restoration of image defects, which are typical for old videotapes; the main partners were the INA (Institut National d'Audiovisuelle, Paris), BBC-London and Snell & Wilcox, UK. PRESTO (Preservation Technology for European Archives) and the subsequent project PRESTOSPACE, within European Commission 6th Framework Programme for Research, provided technical solutions and integrated systems for a complete digital preservation of all kinds of audiovisual collections.
Audiovisual archiving is a complex and multi-disciplinary domain including such diverse fields as chemistry, physics, imaging technology, signal processing, robotics, artificial intelligence and semantic interpretation. These EU-projects bring together participants including archive owners, broadcasters, research centers from archive institutions, general research centers and universities, industries, and international non-profit institutions.
An important result of the movie restoration research is the development of new software: e.g. FRAME and its further development DIAMANT. Several early movies were reconstructed using DIAMANT, the most famous being Metropolis in 2001. LIMELIGHT was a software project that was further developed by the University of La Rochelle to become RETOUCHE.
In the period of 2010-2013, the Swiss National Science Foundation financed the project 'Dust BW: Detection of dust and scratches on photographic silver-halide (black/white) material by polarized dark-field illumination'. Partner of the project were the Imaging & Media Lab of the University of Basel and the Audiovisual Communications Laboratory School of the EPFL (École Polytechnique Fédérale de Lausanne). The present thesis reports the results of this research project.
Several factors are responsible for the deterioration of photographic material; we will only focus on dust and scratches. Different methods have been adopted up to now for the automatic detection of dust and scratches; each method has pros and cons, and a limited field of effectiveness. Infrared radiation (e.g. Digital ICE) and the spatio-temporal image analysis are among the most effective methods, although they have their limits. The infrared radiation only works for dye-based material and the spatio-temporal image analysis is not applicable for still images. The spatio-temporal image analysis is also limited due to motion in the scene and it is not effective for defects appearing in sequences spanning more than one frame (e.g. vertical scratches and lens dust). This thesis presents a comprehensive range of methods for optical flaw detection, applicable on any type of photographic film (both silver-based and dye-based material, as well as still images and moving images). It also presents innovative methods that combine different optical techniques and computational photography.
This thesis begins by describing the structure of a processed photographic film, analyzing its constitutive elements layer by layer. Afterwards, we will provide a list of the most common types of decay affecting the photographic film, classified on the basis of the layer that is primarily affected. This will be followed by a summary of the state-of-the-art strategies and solutions for the restoration of local flaws. The successive analysis of the interaction between light and photographic film provides the scientific framework needed to understand the optical phenomena involved in the experiments. The following chapter explains the phenomena and the techniques involved in imaging photographic film. A specific chapter surveys the optical methods available for flaw detection on any type of photographic transparency, including innovative solutions. The successive chapter defines the procedure we adopted for the numerical evaluation of the performances of the methods analyzed in the experiments. Prior to the conclusions, the thesis finally addresses the experimental analysis, identifying the best parameters for the proposed methods and comparing the innovative methods with the most effective traditional ones.