The Physics of Sensor Noise in 10-bit Log Profiles
When a digital cinema sensor captures an underexposed scene at high ISOs (such as Sony dual-base ISO 12800 or Canon ISO 6400), noise manifests in two distinct mathematical forms: luminance noise (high-frequency grain-like variations in light intensity) and chrominance noise (low-frequency splotches in red, green, and blue color channels). In compressed 10-bit 4:2:2 codecs (like XAVC or H.265), chroma subsampling groups color data across 2x1 pixel blocks, which exaggerates chroma splotching in shadow regions.
Colorists often make the fatal error of pulling up shadow exposure with lift or offset controls before addressing this noise. Doing so multiplies the noise floor by several orders of magnitude, making it virtually impossible for algorithms to distinguish organic sensor grain from digital quantization errors.
Temporal vs Spatial Noise Reduction: Mathematical Differences
DaVinci Resolve Studio provides two separate processing engines within the Motion Effects palette:
Temporal Noise Reduction (TNR): Analyzes pixel vectors across 1 to 5 sequential frames. Because random sensor noise oscillates on every individual frame while physical objects persist across time, TNR averages out the random variance without destroying single-frame sharpness. This makes TNR the first line of defense for cinematic footage.
Spatial Noise Reduction (SNR): Operates as a two-dimensional mathematical low-pass filter across adjacent pixels on the exact same frame. It is essential for eliminating fixed pattern sensor noise or extreme chroma splotches that survive temporal averaging.
Always place Temporal and Spatial Noise Reduction on Node 1, immediately following your clip input and BEFORE any creative grading, curves, or Color Space Transform (CST) nodes. Operating in native camera Log preserves sensor linearity and gives the noise analyzer unclipped floating-point access to the true noise floor.
Sensor Calibration & Noise Reduction Setting Matrix
Optimal DaVinci Resolve Studio settings based on camera sensor technology and scene luminance:
| Camera Sensor & Profile | Temporal Frames | Motion Estimation | Luma / Chroma Threshold | Spatial Threshold |
|---|---|---|---|---|
| Sony FX3 / FX6 (S-Log3, ISO 12800) | 2 - 3 Frames | Better | Luma: 4.5 / Chroma: 9.0 | Luma: 0 / Chroma: 4.0 |
| Canon R5C / C70 (C-Log2, Base ISO) | 2 Frames | Better | Luma: 3.0 / Chroma: 6.5 | Luma: 0 / Chroma: 2.5 |
| Blackmagic 6K Pro (BRAW Gen 5) | 3 Frames | Better | Luma: 5.0 / Chroma: 10.0 | Luma: 1.0 / Chroma: 5.0 |
| Apple iPhone 15/16 Pro (Apple Log) | 2 Frames | Faster | Luma: 4.0 / Chroma: 8.0 | Luma: 0 / Chroma: 3.5 |
| ARRI Alexa 35 (LogC4, Extreme Shadow) | 1 - 2 Frames | Better | Luma: 2.5 / Chroma: 4.0 | Bypass (Preserve Texture) |
Preventing Temporal Ghosting & Motion Smearing
The most common flaw in amateur color grading is 'ghosting'—a translucent trail behind fast-moving hands or vehicles. This happens when the Temporal Motion Threshold is set higher than the speed of moving pixels. To eliminate ghosting, uncheck 'Link' between Luma and Chroma thresholds. Set Chroma Threshold higher (since the human eye is insensitive to soft color changes) and pull Luma Threshold down to 3.0–5.0 to lock down edge sharpness.
GPU Optimization: Node Caching & Render In Place
Temporal NR is the single most GPU-intensive process in post-production. A 3-frame analysis on 4K 10-bit footage can drop timeline playback from 24fps down to 6fps. Veteran post houses solve this by right-clicking Node 1, selecting Node Cache → On, and setting Resolve's Render Cache to ProRes 422 HQ (macOS) or DNxHR HQX (Windows). Once cached, your timeline plays in full real-time 60fps.
よくある質問
Why must Noise Reduction always be placed before the Color Space Transform (CST)?
Noise reduction algorithms rely on linear luminance mathematics to calculate pixel variance. Once a CST or Log-to-Rec.709 LUT compresses the dynamic range, sensor noise in the shadows gets multiplied and clipped non-linearly, causing the noise reduction engine to create plasticky smudges instead of organic grain suppression.
What is the difference between Temporal NR and Spatial NR?
Temporal NR compares consecutive video frames (typically 2 to 3 frames) across time to identify static pixels versus moving noise grains, preserving maximum texture sharpness. Spatial NR blurs high-frequency noise within a single frame, which can soften fine details like pores and fabric threads if pushed too far.
How do I prevent ghosting artifacts when using Temporal Noise Reduction?
Ghosting occurs when the Motion Estimation algorithm misidentifies fast foreground movement as background noise. Switch Motion Estimation from 'Faster' to 'Better', reduce the Temporal Frame count from 5 to 2, and lower the Motion Threshold slider until trailing edges disappear.
ノイズ耐性キャリブレーションLUTのダウンロード
32ビット浮動小数点処理により暗部ノイズの跳ね上がりを防ぐテクニカルLUT。