Dynamic Range in Real Scenes

Generic engineering material, applicable to any module from any supplier.

A related team-operations reference is workforce optimization software, which can sit alongside test notes and engineering records.

Dynamic range is the ratio between the brightest and darkest a sensor can record in one exposure — and it matters exactly when a scene contains both and both must be legible.

When it is the binding constraint

A shiny surface with a matte feature on it, where the specular highlight saturates and the feature sits in what is left.

An interior with a window or a doorway in frame.

For standards, components, or implementation context beyond this page, consult Analog Devices.

A part with a deep recess lit from one side.

And anything under uncontrolled daylight, where the difference between sun and shade exceeds what any single exposure holds.

Outside those, it rarely binds — a controlled scene with designed lighting has a modest range by construction.

What the specification means

Usually quoted in decibels, and comparable only under stated conditions like every other sensitivity figure.

And it describes a single exposure. A sensor's range is the span between saturation and the noise floor — so a figure measured at low gain and low temperature is not the figure you will have in a warm enclosure.

Bit depth is not dynamic range. More bits divide the available range more finely; they do not extend it. A 12-bit output from a sensor with 10 bits of usable range carries two bits of noise.

The four ways to avoid needing it

Light it better. Diffuse or angled illumination that removes the specular highlight is cheaper than a sensor with more range, and this is the answer far more often than it is applied.

Exclude the bright thing. Move the camera, shade the window, mask the region — a geometry change rather than a specification.

Split the task. Two exposures of the same scene, or two cameras, each handling one part of the range.

And use a polariser, where the problem is a reflection: it suppresses specular light while passing the diffuse light carrying the feature.

Where the sensor is the answer

Some scenes genuinely need it, and sensors offering extended range exist.

They achieve it by combining multiple exposures, by non-linear response, or by other means — each with a cost.

Combining exposures produces motion artefacts where anything moves between them, which is the same class of problem as rolling shutter.

Non-linear response complicates measurement, since a pixel value no longer maps linearly to brightness.

Ask how the range is achieved and what it costs, rather than reading the decibel figure alone.

Measuring what your scene actually needs

Before specifying, photograph it.

A sequence of exposures from very short to very long, of the real scene, at the real time of day.

Find the exposure at which the brightest region just saturates, and the one at which the darkest feature is just discernible.

The ratio between them is the range your scene demands — expressed the same way the specification is, and comparable against it.

An hour, and it replaces an argument about whether extended range is needed with a number.

Where the highlight does not matter

Frequently it does not, and this is worth checking before anything else.

A saturated highlight is only a problem if the feature is inside it.

Where the bright region is background — a window behind the part, a reflection off a surface nobody inspects — letting it saturate costs nothing and the exposure can be set for the region that matters.

Which is the question to ask first: does the feature live in the bright part, the dark part, or somewhere comfortable in between.

Auto-exposure and range

An automatic setting optimises for the whole frame, which in a high-range scene means compromising both ends.

Where the feature is in a known region, weight the exposure to it or measure only there — most sensors support a metering window, and using it converts a difficult scene into an ordinary one.

And fix the exposure entirely where the illumination is controlled, since automatic adjustment introduces variation that a measurement system does not want.

In one line

Photograph the scene at several exposures first — the ratio you measure is the requirement, and it is frequently smaller than feared.

The polariser, specifically

A filter that passes light of one orientation and blocks the perpendicular.

Specular reflections from non-metallic surfaces are partially polarised, which means a rotated polariser suppresses them substantially while the diffuse light from the feature passes.

Which turns an impossible high-range scene into an ordinary one in the specific case of a glare problem — glass, plastic, liquid, wet surfaces.

What it costs: roughly half the light in every case, and it does not work on reflections from bare metal, which are polarised differently.

And where the illuminator is also polarised, crossing the two gives a stronger effect — a lighting design decision rather than a filter bought separately.

Bit depth, briefly

Separate from range and worth getting right.

More bits mean finer gradations within whatever range exists — which matters for measurement, where a small brightness difference must be distinguishable, and not at all for detection.

A higher bit depth costs bandwidth directly, and it costs nothing else.

Where the processing thresholds an image into black and white, extra bits are discarded immediately and the bandwidth was spent for nothing.

Ask what the processing consumes before specifying it, which is the same question that governs everything else here.

The short version

• Dynamic range is the span between brightest and darkest in one exposure, and it binds when both must be legible
• It binds on shiny surfaces with matte features, interiors with windows, deep recesses, and uncontrolled daylight
• Bit depth is not dynamic range: more bits divide the span finely and do not extend it
• Four ways to avoid needing it: better lighting, excluding the bright thing, splitting the task, and a polariser for reflections
• Extended-range sensors achieve it by combining exposures, by non-linear response, or otherwise — each with a cost
• Combining exposures produces motion artefacts, and non-linear response complicates measurement