ReCAP for real-time detection of colour bleeding artefacts

ReCAP for real-time detection of colour bleeding artefacts in chroma key scenes addresses a subtle but visible problem in composited video. When keyed edges retain background colour, green-screen spill creates halos around hair, clothing and shoulders, while aggressive keying can leave colour contamination inside the subject. Broadcast-quality analysis needs to identify these defects consistently, locate them in the frame and preserve evidence for operators.

For Australian production teams, the challenge is magnified by live sport, rapidly produced studio segments and outdoor shoots where daylight and weather vary between setups. A detector must work across different cameras, codecs and key colours without treating every soft edge as a fault. ReCAP’s broader capabilities in metadata extraction, quality monitoring, logo and face recognition, and duplicate detection provide a foundation for combining visual defects with production context.

Why edge contamination matters in Australian production

A clean key should separate the subject from its replacement background without altering natural skin tones or costume colours. Edge contamination occurs when the keyer removes too little background colour, removes too much subject detail, or allows the chroma plane to spread into neighbouring pixels during encoding. The visible result may be a green fringe, a coloured outline, patchy transparency or an apparently clean silhouette that becomes defective when scaled for broadcast.

Australian productions encounter many conditions that can expose weak matte handling. Cricket footage often places players against large green areas, making it difficult to distinguish the playing surface from similarly coloured clothing or equipment. Exterior scenes around Sydney and Melbourne may move between overcast interiors and strong daylight, causing camera colour responses to change between takes. Live news and studio broadcasts also have little opportunity to correct a persistent halo after transmission.

The business impact is greater than an isolated visual blemish. A recurring fringe can make a presenter appear poorly made-up, distract viewers from a performer or undermine the perceived quality of a commercial. ReCAP can treat edge artefacts as measurable quality events rather than subjective notes. Frames can be linked to timecode, camera metadata, scene boundaries and codec conditions, giving editors and broadcast engineers a common record of where the issue occurred.

From chroma spill to real-time quality evidence

A practical detection method examines the boundary where the keyed foreground meets the replacement background. For each candidate edge, the system can compare the subject and background chroma values, estimate matte softness and measure the width of the transition band. This reveals both coloured spill outside the subject and contamination extending inward from the edge. The analysis can be performed on decoded frames or on a dedicated chroma representation where the workflow provides one.

Compression must be part of that assessment. Chroma subsampling can blur colour across an edge, while bitrate limits can create blocks or transient shifts that resemble a defective key. Temporal behaviour is therefore as important as a single-frame score. A genuine keying artefact often persists or moves consistently with the subject, whereas compression noise may fluctuate independently. A ReCAP-based workflow can combine edge distance, colour similarity, spatial extent and temporal persistence into a confidence score.

Real-time processing does not require every pixel to receive the same scrutiny. A lightweight stage can screen likely regions, after which a more detailed model evaluates the strongest candidates. This reduces computation while retaining enough context to distinguish a narrow halo from a deliberately coloured costume. The resulting event should include a bounding region, severity, affected time range and the signal that triggered it, allowing an operator to review the source rather than simply accept a binary pass or fail decision.

Reference bars, metadata and traceable logs

Reference patterns help distinguish a production fault from an expected signal. ReCAP has documented an incorrect colour-bar logging method, which offers a useful model for recording objective media irregularities. Colour bars do not themselves reveal green-screen spill, but the same principle applies: detection needs traceable measurements, consistent thresholds and an audit trail that survives beyond the original monitoring session.

For chroma-key analysis, each event should be connected to source and timecode identifiers, resolution, frame rate, camera or feed details, codec, key colour and processing status. A catalogue can then show whether contamination appears mainly in one camera, after a particular grade, or only within a compressed distribution stream. That context is essential when multiple feeds share a workflow or when a network delivery path introduces errors that were absent at capture.

Logs should also preserve separation between observed data and interpretation. The detector can report that a green-dominant band appears 12 pixels wide along a subject’s shoulder, but it should not assume that every green-dominant edge is defective. Reviewers can classify confirmed defects, reject false positives and attach notes. These decisions improve threshold tuning while retaining the original measurements, making the system useful for quality assurance rather than merely an automatic warning light.

Live workflows that fit Australian teams

The most immediate use case is a live gallery or vision-mixer environment. If ReCAP identifies a persistent coloured halo, the operator can check key settings, background sources and the health of the composited feed before the artefact reaches air. In a pre-recorded workflow, the same event can be sent to an editorial or post-production queue with a clip around the detected interval. This shared evidence reduces the need for producers, editors and engineers to describe the same fault in different terms.

Media asset management is another natural fit. A production company in Melbourne or Sydney may handle daily news, sport, studio programmes and advertising material with different keying standards. Integrating the detector with an existing asset catalogue lets teams compare scenes, identify repeated issues and determine whether a faulty matte originated in capture, post-production or delivery. ReCAP’s other analysis functions can add useful context, including detected logos, faces and duplicated sequences, without making the colour analysis dependent on identity recognition.

Network conditions must be tested before operational use. As a browser performance check, engineers can time the dashboard and a public web interface sample, although only ReCAP measurements belong in the quality record. A useful pilot measures detection latency, review-queue stability and the behaviour of the alert interface under a busy workload. The output should remain accessible during a live transmission, while full-resolution analysis and forensic logging can continue in the background.

Validation limits and operational safeguards

No automatic detector can interpret every chroma-key scene without a controlled error rate. Costumes, reflective surfaces, motion blur, translucent fabric and coloured practical lights can all resemble contamination. Test material should therefore include Australian presenters, darker skin tones, grey and white hair, saturated uniforms, transparent objects and fast movement. Results should be reviewed by people familiar with the intended visual treatment, especially when the workflow combines chroma analysis with face recognition or other biometric processing.

Thresholds should be configured by workflow rather than imposed as universal values. A live sports composite may tolerate a narrow transition where a premium drama sequence demands a more exact matte. Colour-management settings, camera colour science and delivery codecs also influence the signals, so a baseline needs to be established for each production profile. Human reviewers should be able to mark a flag as genuine, irrelevant or uncertain, with those outcomes used to adjust the model without erasing conflicting evidence.

As an EU-funded research initiative, ReCAP also needs transparent governance around stored footage, access permissions and retention. Logs should contain the minimum information needed for quality review, and any face-related output should be handled separately from pixel-level defect evidence. These safeguards make the results more defensible and prevent a technical score from being treated as a judgement about a person or performance.

A deployment can begin with a small, representative test set drawn from green-screen material used by the production team. ReCAP should return an edge map, confidence score, severity rating and source-linked log entry for each flagged event. The next step is to record one minute of green-screen footage with a deliberately soft edge and compression, run it through the detector, and compare every flagged frame with a human matte review.