Digital Film Restoration
For the field of digital film restoration, we have made various developments for direct production use and have been working according to this manner since 2008. We use an open pipeline for the entire process, extended with various tools of our own. This makes us flexible and at the same time efficient in solving new challenges.
In the field of film scanning, we prefer to work on discrete single-frame sequences. Classically, these are mostly DPX or TIF files. In our pipeline, we generally use lossless compression. As a result, the project automatically receives image-based checksums, which inherently ensure the integrity of the data. In addition, we save significant storage space and can work without unnecessary chroma subsampling.
Since the process of digitalization is a basis for each subsequent step of the workflow it has great importance. Therefore, it must receive the best possible quality.
Whether or not a gentle cleaning makes sense, is based on the review results. Depending on the quality of the source material, we often recommend 4K, 16bit (for example, 4096 x 3112 pixels) as the scan resolution.
As the development process of a dup-positive is able to normalize the grading of a film, the first step during and after scanning is the determination of light. During the scanning, the light intensities can be used to target weaknesses or the character of the material. We recommend based on scenes or individual shots. This measure of adjusting the scanning light clearly increases the room for correction of digital processing.
Light Determination prepares the material for retouching, so the Color/Light-Appearance is already set at this time.
Contrary to common approaches, where the frame is defined based on a setting for an entire act of a movie, we work with an internal system that automatically recognizes and sets the largest possible active image area for each scene (shot) between two cuts.
Our auto-framing is stable in different scenarios and has already been tested several times in practice. An immenent advantage lies in the linkability with a stabilization. Also in this development, our focus was on the highest possible processing quality. This combination is standard in this form only with us.
- individual determination of the best possible (largest) image section
- fully automatic with many options for manual adjustment
- stable even in dark scenes, in ones with strong image movement or severe damage such as cracks etc.
- applied in combination with stabilization only once a sampling of images is done (an important quality factor)
For stabilization combined with auto-framing , the high-frequency shift between the images is compensated based on the image content. Both operations (auto-framing and stabilization) are combined and result in one sampling process (shift / rotation / scaling) of the image.
Special attention in this fully automatic process is given to the analysis of a film. This data is handed, on shot basis, to all upcoming actions in the pipeline. As a result, corrections can be introduced individually, for example by weakening the stabilization for short passages in their influence or manually stabilizing them.
In addition to the technical benefits of flexibility and quality, problems with foreground dominance can be easily resolved. For us, this criterion is always important because other systems (software/hardware) often produce this not so easily recognizable artifact. It becomes visible through movement which was not part of a scene before i.e. when faces/people occupying much of the foreground and there is little contrast in the background.
To remove dust, dirt and scratches, we use a multi-level approach to motion estimation and compensation that eventually fills appropriate content into detected error masks. As a result, the retouching quality of larger defects is stable and working in high quality .
Individual configuration of the features:
- Temporal Dirt
- max Diff to Global Motion
- robust larger Dirt
- adaptive enhanced Dirt Detection
- adjacent Content Reliance
- Spatial Dirt
Density variations are a typical effect that is primarily caused by deviating exposure times during film recording or copying. To significantly reduce this effect, we use a combination of global and local correction within an automatic and adaptive approach.
Adaptive limits are helpful for avoiding unwanted irregularities such as explosions visible in the image, lightning or strong light reflexes.
Based on silent film projects, we have developed a system of unwarping based on the image content. Distortions are caused by different stresses in the material during storage or by the layer-based composition itself.
The distortions occurring after slices (splices) can also be selectively removed with high-quality.
- Equalization of each scene based on static camera content or nodal pan
- Equalization based on surrounding neighbor images in case of disturbances (i.e. at splices)
Lady Hamilton – DeWarping, nodal pan
Lady Hamilton – DeWarping, fade e.a.
In addition to material heavily loaded with defects, the beginning and end of individual acts can be severely burdened with disturbances which not only occasionally occlude the pictorial content of the individual images, but also sometimes mixed at texture level.
Our system is adaptive and robust enough to perform this kind of retouching step. These are complemented by different DeNoise methods.
Based on the films “The Blue Straw Hat” and “Riviera Story”, we have developed a retouching system for bacteria. Here we are talking about pictures, of which at least 10 to 15% of their content area is completely obscured by defects.
The technical challenge here is both the stable detection and retouching of these disturbances when there is a defect in the same place in successive images. This retouching is not a standard and was developed specifically for this purpose and adapted further. We adjusted a special reliance method to our DustBusting methods, which are derived from experience out of this project.
“Der blaue Strohhut”, automatic (no manual correction)
Robust detection in comparison (other DISR-Systems)
Based on the requirements of Fritz Lang’s Metropolis 27/10, we have enhanced our workflows with concepts for preserving or restoring grain (film grain). With this approach it is possible to combine different film materials, in this special case 35mm original negative and found 16mm positive copies.
In current workflows, the texture management ensures the retouching without standing grain, so that inserted pieces are not structured identically, as in the origin image.
In other cases, it is also noise emulation from CCD or CMOS sensors.
Our focus is in the visual world. Nevertheless, we have developed some automations for Picture + Sound and always apply them whenever appropriate. For services beyond the following focal points, we cooperate with competent local partners in Berlin.
- Sample rate conversions
- Adjusting the playback time to frame rate
- Sequential filtering for known weaknesses (DeHum etc.)