No-reference measurement
Perceptual Analysis
Measure decoded-content quality without a pristine reference. Get a 0–100 relative quality score with grade, confidence, and metric evidence.
Published result contract
The 0–100 score is a relative quality index. Review its grade, confidence, coverage, and result state together.
Details: Published result contract
- Result
- Perceptual Quality Score, 0-100
- Score type
- Relative objective quality index
- Default mode
- Balanced
- Default profile
- Webcam
- Grade thresholds
- Excellent 85, Good 70, Fair 55, Poor 40, Bad below 40
- Configuration
- Explicit profile, source class, style, lighting, use, and sampling
- Result states
- Complete, provisional, partial, spatial-only, legacy, unavailable, invalid
The score is a relative objective quality index. Confidence measures whether evidence is suitable and sufficiently covered; it does not change a primary metric's measured quality or direct contribution, but it can affect supporting deductions, evidence coverage, and grade eligibility. A provisional result has a published score but does not satisfy every evidence requirement for a complete result.
Scoring evidence
Naturalness, detail, cleanliness, and perceptual artifacts form the primary evidence. Supporting defects can deduct points within a bounded budget.
Details: Scoring evidence
| Evidence | Role | Technical interpretation |
|---|---|---|
| Visual Naturalness | Primary | NIQE natural-scene-statistics distance using the validated pristine model. Lower native measurements map to higher normalized quality. |
| Sharpness & Detail | Primary | Multiscale tile-pooled Laplacian energy with clipping/dark-tile rejection, noise prefiltering, motion-conditioned applicability, and separate natural/edited pooling. |
| Image Cleanliness | Primary | Percentage of eligible low-structure luma pixels above the median-residual threshold. Chroma sigma and motion-compensated temporal residuals are retained as diagnostics rather than replacing the published measurement. |
| Perceptual Artifacts | Primary | BRISQUE-based distortion evidence using the validated LIVE model/range pair. |
| Compression Artifacts | Supporting | Multiscale periodic boundary discontinuity with phase search and strong natural-edge suppression. |
| Detail Structure | Supporting | Broad healthy edge-density band; cannot reward unbounded edge density. |
| Tonal Complexity | Supporting | Broad entropy band with high-side noise protection. |
| Color Richness | Supporting | Broad plateau that detects color collapse without rewarding excessive diversity. |
Primary evidence forms the core score. Supporting evidence cannot add positive points; defects consume a bounded penalty budget.
Scene, motion, and edited-content handling
Scene transitions, edited content, and motion affect which evidence applies and how it is pooled.
Details: Scene, motion, and edited-content handling
Temporal pooling
Valid frame observations are normalized before pooling. Natural scenes use a weighted median/P20 blend; edited or mixed content uses a more median-dominant pool. Scene results are combined by retained evidence weight.
Transition exclusion
Hard cuts, fades, and dissolves use type-specific exclusion windows so transition frames are retained in provenance but excluded from defect pooling.
Edited content
Sustained scene-boundary frequency and graphic-dominant samples can reduce naturalness influence and move released weight toward persistent detail and cleanliness.
Motion context
Optical-flow activity and coherence condition sharpness applicability and temporal analysis. Motion is a descriptor; high activity does not receive quality points.
Studio evidence and export
Inspect scores and metric contributions in Studio. Licensed builds retain ranges, raw evidence, and provenance in CSV and JSON.
Details: Studio evidence and export
On screen
- Input queue, sampled-frame preview, and finite-video range selection
- Analyze/Pause/Resume with a separate Stop control
- Overall score, grade, confidence, and evidence coverage
- Metric type, normalized quality, configured weight or penalty budget
CSV and JSON
- Source identity, codec, resolution, bitrate, and configuration
- Inclusive frame bounds and corresponding positions in seconds
- Raw metrics, samples, scenes, transition exclusions, and partial status
- Model identity and measurement provenance in structured JSON
For finite video, the selected inclusive frame range is applied before the sampling interval. Pause takes effect at the next frame boundary, Resume continues the same run, and Stop retains completed evidence as a partial result. Inputs and range controls remain locked while analysis is active or paused.