Photography Lighting & Exposure
Dynamic Range Recovery in 14-Bit RAW Files Compared to In-Camera JPEG
A camera can show a white patch as blown out in its JPEG while the paired RAW still holds the weave. The reverse problem appears in shadows, where a clean-looking JPEG may have smoothed away detail. Use one-third-stop brackets, channel readings, and repeated shadow frames to measure the remaining highlight margin and set a usable shadow limit for your lighting setup.
What the code values measure
A 14-bit RAW sample can represent 16,384 nominal levels at a sensor site. In a linear RAW readout, equal code increments represent equal increments of signal, up to saturation. An in-camera 8-bit JPEG decodes to 256 possible values per RGB channel after demosaicing, white balance, color conversion, tone mapping, and lossy compression. The camera’s tone curve assigns those codes unevenly across scene exposures. The sRGB transfer function adds another nonlinear mapping, so a change from code 16 to 32 carries a different linear-light increment from a change from 192 to 208. W3C sRGB specification
Suppose your RAW analyzer reports a black level of 512 and a white level of 15,872. That leaves a black-subtracted span of 15,360 counts. The brightest stop spans roughly 7,680 counts, while the next spans 3,840. A signal eight stops below saturation sits at 60 counts above black. Doubling it to 120 in a linear edit also doubles any noise already present. The nominal bit count describes the precision of the encoding. The saturation point and noise determine how much of that precision is useful. Uncompressed storage preserves the recorded samples, and lossless compression preserves those same samples when the camera uses the same readout mode.
Build a repeatable lighting test
Mount the camera on a tripod and place a matte white textile, neutral gray card, and dark textured cloth in the same plane. Add a small matte red swatch to reveal early clipping in a saturated color. Set a stable continuous key one metre from the middle of the target, 45 degrees off the camera axis as seen from above. Keep the key position fixed and eliminate changing ambient light. Check that the white weave resolves clearly in a safely exposed frame and that the swatches receive consistent light.
At base ISO, select manual exposure, a fixed aperture such as f/8, manual focus, fixed white balance, and simultaneous uncompressed 14-bit RAW plus highest-quality JPEG. Document the JPEG profile and noise reduction setting. Switch off automatic contrast and highlight modes for this run. Check the camera menu and file metadata for the chosen RAW mode, since some shutter or burst settings change the readout precision. With the gray card as your exposure reference, choose a baseline where the bright textile still separates in both files.
Bracket with shutter speed in one-third-stop increments from the baseline until both formats clip. Then bracket down in one-stop increments to at least −5 stops for the dark cloth test, taking two frames per setting. Use shutter times that give repeatable gray-card readings under your lamp. If two exposures at the same setting disagree or show stripes, resolve flicker before interpreting the bracket. With flash, repeat the setup using measured changes in flash output at a fixed sync-safe shutter speed.
Find the last recoverable highlight
Sample the same interior crop of the white textile in each file. In the JPEG, inspect decoded RGB channel values and the weave together. For this test, call a channel clipped when a contiguous area sits at codes 254–255 and the finer tonal variation visible in a lower bracket disappears there. Check the red swatch separately because color can shift before the white target fails. Use a RAW analyzer that displays pre-demosaic samples by color channel and reports the applicable white levels. Flag the first repeatable plateau at the white level in a channel across the crop, excluding isolated hot pixels or glints.
In the illustrative bracket, the JPEG first meets the clipping criterion at +⅓ stop. The RAW still retains all channels and weave at +⅔ and first clips at +1. That gives a two-step gap of ⅔ stop between the first clipped frames, and +⅔ stop of verified RAW headroom above baseline for this white patch. The exact clipping thresholds lie between sampled exposures. If the red patch fails earlier, use its limit when red color fidelity matters.
To check actual recovery, normalize both versions of the +⅓-stop frame toward the baseline gray card and compare the same weave at identical size. A clipped JPEG patch can turn gray when darkened while remaining flat. In a RAW conversion, switch off highlight reconstruction for the diagnostic pass. Re-enable it separately to judge whether an intact channel can supply useful texture when another clips. Reconstructed color is an estimate, and an area saturated across all contributing channels lacks recorded highlight texture. Adobe Camera Raw tonal controls
Measure the shadow floor and check texture
Capture a pair of capped frames at the ISO and shutter time you are testing. In a central crop, compare corresponding samples from a single RAW color plane. Subtract frame B from frame A, then divide the standard deviation of the differences by √2 to estimate temporal dark noise in RAW counts. Use matching pairs for other shutter times if they differ substantially. Frame subtraction cancels stable patterns, so inspect individual frames for banding and uneven black levels too. EMVA 1288 linear camera measurement standard
Repeat the pair calculation in a smooth dark patch beside the cloth to include photon noise. In this example the capped pair gives 4 counts of temporal noise. The −3-stop patch averages 48 counts above black with 7 counts of temporal noise, for a signal-to-noise ratio of about 6.9:1. At −4 stops, 24 counts with 6 counts of noise gives 4:1. With an illustrative acceptance threshold of 5:1, −3 passes the noise check and −4 fails. Inspect the cloth separately because a flat patch can pass while the weave still disappears under processing.
Match both formats to the baseline gray brightness in a high-precision editor and view at the intended delivery size. Keep RAW denoising and sharpening off for the diagnostic pass. A JPEG may show less scatter because its processing smooths neighboring tones. Look for identical threads across repeated frames, along with colored blotches, flattened grain, or block edges. Record the deepest bracket that keeps the same recognizable texture and acceptable color in each format. Rendered JPEG code scatter measures the processed image and has different units from RAW sensor noise.
Use the measurements on set
| Check | Uncompressed 14-bit RAW | In-camera 8-bit JPEG |
|---|---|---|
| Quantization | 16,384 nominal codes per sensor sample | 256 decoded values per RGB channel after rendering |
| Highlight limit | Check RAW channel saturation and retained texture | Check channel plateaus and lost texture in the chosen profile |
| Shadow limit | Measure temporal noise and inspect lifted texture | Inspect lifted texture, color, and compression artifacts |
If the white patch clips in JPEG at +⅓ and RAW at +1, the baseline is the last verified JPEG frame and +⅔ is the last verified RAW frame. For a setup that varies by roughly one bracket step between frames, back off another third from the last passing exposure. If that puts the cloth below your shadow limit, add controlled fill and measure again. Doubling the total illumination on the dark cloth raises its exposure one stop. Flag the fill off the white textile so the highlight margin survives, then check both patches after the change.
Verify before locking exposure
- Record ISO, shutter mode, RAW precision, JPEG profile, white balance, and the key’s distance and angle.
- Log the last passing and first clipped bracket for the white textile and the critical color swatch in both formats.
- Retain the paired frames and note the RAW noise estimate, the shadow texture limit for each format, and the delivery size used for inspection.
- Make one final pair at the production exposure after adding fill or moving a flag, then recheck the brightest channel and darkest texture that matter.