Using ReCAP to Log Color Palette Shifts in Film Restoration
Film restoration is often described as a search for the image a production originally intended to show. In practice, that image may have changed through ageing film stock, repeated transfers, damaged negatives, analogue grading, television mastering and digital compression. A restoration team therefore needs more than a visually pleasing result. It needs a traceable record of how colour, contrast and tone behave across a feature, episode or archive collection.
ReCAP, the EU-funded Real-time Content Analysis and Processing project, provides a useful framework for this work. Its focus on broadcast-quality video analysis, metadata extraction, quality monitoring, face and logo recognition, and duplicate-content detection can be extended into a structured method for observing colour palette shifts. The result is a searchable restoration log that connects visual evidence with reels, shots, timecodes, versions and delivery files.
Why Colour Changes Matter During Restoration
Colour shifts can be caused by the source material, the scanning process or a later mastering decision. A faded print may lose cyan and retain warmer red tones, while a poorly stored negative can show uneven density from one section to another. A telecine transfer may introduce a different white balance, and a streaming encode can alter shadow detail without changing the creative grade. These changes are easy to notice in a side-by-side comparison, but harder to track consistently across many hours of footage.
A palette log turns subjective observations into reviewable evidence. It can record dominant hues, saturation ranges, luminance distribution, skin-tone tendencies and abrupt changes between adjacent shots. The information does not replace a colourist’s judgement. Instead, it helps identify where judgement is required and makes the reasoning behind a restoration easier to revisit.
This is especially valuable when the available references disagree. A production may survive as a camera negative, a release print, a broadcast master and a compressed online copy. None should automatically be treated as definitive. A documented analysis can show that a green cast appears in every surviving source, or that it begins only after a particular reel change. That distinction can prevent an authentic visual characteristic from being “corrected” out of the picture.
Building A ReCAP Analysis Workflow
A practical workflow begins with ingest and normalisation. Each file should receive stable identifiers for title, episode, reel, source format, frame rate, resolution, scan date and storage location. ReCAP’s metadata capabilities can support this catalogue, while quality-analysis functions can flag damaged frames, dropped frames, excessive noise, clipping and other conditions that may influence colour measurements.
The next stage is shot-aware sampling. Rather than calculating one average colour value for an entire film, the system should examine representative frames or short intervals within each shot. A single frame may contain a bright window, a fire or a costume that distorts the result. Sampling several frames and combining the measurements produces a more reliable description of the shot’s visual palette.
Colour analysis should be performed in a suitable colour space and under controlled technical conditions. RGB values alone are not always perceptually meaningful, especially when comparing different transfers. A workflow may use HSV, HSL, Lab or another perceptual model to estimate hue movement, saturation changes and lightness variation. The selected space, transfer function, bit depth and white point should be written into the processing metadata so later analysts understand how the values were produced.
ReCAP can act as the coordination layer for these events: source detected, quality issue identified, shot boundary located, palette measurement generated and review note attached. If a dedicated colour-extraction module is not part of a particular deployment, it can be added as a processing component that sends results into the same metadata and monitoring pipeline.
Signals Worth Recording
A useful palette log should combine numerical indicators with descriptive labels. Numerical values make it possible to compare shots and versions, while labels help restoration staff interpret what the change looks like on screen. The log should preserve the original measurement as well as any human annotation, rather than replacing one with the other.
The following fields provide a compact starting point for a film or television restoration:
- Dominant hue families and their relative proportions
- Average and percentile luminance across sampled frames
- Saturation, colour variance and shadow or highlight clipping
- Timecode, reel, shot identifier and source-file reference
Abrupt variation is often more informative than an overall average. A gradual warm drift across a reel may indicate fading or a scanner calibration issue, while a sharp blue shift at a splice may point to a source change. A sequence can also contain deliberate palette design, such as a move from muted interiors to highly saturated dream imagery. Automated detection should therefore identify transitions for review, not label every transition as damage.
Thresholds need to be set with care. A 5 per cent change in saturation may be significant in a restrained black-and-white production but insignificant in a music video filled with coloured lighting. It is sensible to establish thresholds by title, source type and section of the programme. The system can then produce alerts when a shot departs from the local visual baseline, rather than applying one universal rule to every archive.
Separating Restoration Evidence From Creative Decisions
Colour-palette extraction is an evidentiary tool, not an automatic grading instruction. A shift detected by ReCAP may reflect an original creative choice, an optical effect, a lighting change or a genuine technical fault. Restoration teams should compare the alert with adjacent shots, production notes, surviving prints, laboratory reports and interviews where available.
Face and logo recognition can add useful context. If a detected palette change coincides with a studio ident, archival insert or television commercial embedded in a programme, it may belong to a different production layer. Face tracking can help group recurring characters and reveal whether skin tones change only for one performer, one location or one source segment. These features should be used with appropriate safeguards, particularly when working with identifiable people.
Duplicate-content detection also supports version comparison. The same scene may appear in a theatrical master, a censored broadcast edit and a restored reference file with different timing or grading. Matching repeated content lets the team compare equivalent shots instead of relying on file names. A log can then state that a warm cast occurs in all versions, or that it appears only in a particular broadcast copy.
Human review remains essential. A colourist, archivist or restoration supervisor should be able to accept, reject or qualify an automated observation. Each decision should include the reviewer, date, source version and reason. This creates a defensible audit trail and avoids the common problem of undocumented “small fixes” accumulating throughout a long project.
Applying The Method In Australia
Australian restoration work often involves material distributed across state institutions, broadcasters, production companies and private collections. A project based in Sydney may receive film elements from Melbourne, a broadcast master from Brisbane and reference files held in Perth or Adelaide. ReCAP’s centralised metadata approach can connect those sources without assuming they share the same naming conventions, scanners or delivery specifications.
The local market also spans feature films, television drama, documentaries, sports archives, advertising and Indigenous screen collections. Some projects are destined for cinema, while others must meet broadcaster, education, museum or streaming requirements. Analysis should therefore preserve the distinctions between a restoration master, a broadcast version and an access copy. A palette shift introduced for a particular delivery format should not accidentally become the archival reference.
Australian viewing conditions can influence review practice. Bright daylight in a home office, large shared screens in production facilities and routine work across different time zones can make informal visual comparisons inconsistent. Teams should use calibrated displays where possible, record display and monitoring conditions, and rely on measurement-backed reports when collaborators in Sydney, Melbourne and regional locations are assessing the same material.
A deployment can organise local operating requirements into a concise checklist:
- Confirm rights, source ownership and permitted processing before ingest
- Record scanner, display, codec, colour-management and delivery details
- Apply Australian privacy and archival-access rules to recognised faces
- Preserve source files and analysis logs in an auditable storage system
The Privacy Act 1988 may be relevant when face-recognition outputs or other personal information are processed, particularly for contemporary footage or collections containing living people. The Copyright Act 1968 also matters when making working copies, extracting frames or supplying review files to external contractors. Rights and privacy assessments should be handled by the responsible organisation, with ReCAP logs storing only the information necessary for the restoration task.
Projects funded or supported through Australia’s screen sector may also have reporting expectations around access, provenance and preservation. Clear metadata assists conversations with institutions such as the National Film and Sound Archive of Australia, state collections, broadcasters and rights holders, even when those organisations use different systems. The benefit is practical: a restoration team can explain what was measured, which source was preferred and why a particular colour decision was made.
Turning Logs Into A Reliable Restoration Record
The most useful output is a layered report rather than a single score. At title level, it might show the overall palette distribution and the number of flagged transitions. At reel or episode level, it can map changes against source material and quality events. At shot level, it should link the measurement to a frame range, preview image, technical note and review decision.
Visual dashboards can make this information accessible to both technical and editorial staff. A timeline may show saturation, luminance and hue movement alongside detected cuts, logos, faces and duplicate segments. Selecting an alert should open the relevant frame and source reference. This design keeps the evidence close to the image, reducing the need to search through separate spreadsheets and folders.
Version control is essential once grading begins. The original scan, restoration intermediate, approved master and delivery encodes should each have distinct identifiers. If a colourist adjusts a sequence after reviewing a ReCAP alert, the project should retain both the pre-grade measurement and the post-grade measurement. That comparison demonstrates whether the change reduced an unwanted deviation without erasing the source evidence.
A strong log also supports future re-use. If a broadcaster requests a new version for Australian television, or a cinematheque prepares a different projection master, the team can return to the measured source rather than beginning from memory. The same framework can support catalogue search, quality assurance and comparative research across a director’s work or a studio’s archive.
The practical takeaway is to use ReCAP as a traceability layer around colour analysis: preserve the source, measure palette behaviour by shot, connect every alert to technical metadata, and require informed human review before changing the image. This approach gives restoration teams a clearer basis for decisions while protecting the history contained in the original footage.