AI Glossary / Computer vision and multimodal AI / Beginner

HDR (high dynamic range)

Definition

HDR, or high dynamic range, describes imaging representations or systems that handle a broad range between dark and bright values. It can refer to scene capture, image data, processing or display capabilities, so the specific stage and measurement conditions must be stated.

What it means

A scene containing a dim interior and a bright window can have a larger brightness range than one exposure or display can reproduce faithfully. HDR capture and representation aim to retain information across more of that range. HDR display concerns showing a wider range of luminance under specified conditions. Tone mapping adapts an image to a destination's capabilities and can be used for either HDR or conventional outputs. The familiar strongly processed photographic look is only one possible style; HDR itself does not mean exaggerated color, halos or uniformly bright shadows.

How it works

A capture system may obtain broad-range information through a suitable sensor, multiple exposures or other acquisition techniques. The data must use an encoding capable of representing the intended range and precision. Processing preserves or modifies those values according to the task, then a display or output transform maps them to a reproducible appearance.
Dynamic range is often expressed as a ratio or in stops, with one stop representing a factor of two. The relevant limits depend on context: sensor saturation and noise, scene radiance or display black and peak luminance are different quantities. File bit depth alone does not establish the complete dynamic range of a capture or display system.

Three practical examples

• An architectural photographer captures both a room interior and a sunlit window view, then produces an output suited to the publication medium.
• A visual-effects team uses a scene-referred environment representation with bright light-source values to support rendering and lighting calculations.
• A video-production team prepares material for an HDR delivery workflow and checks its appearance on a suitable, configured display rather than assuming any screen shows the same result.

Think of it this way

Think of a measuring instrument with a wider usable scale, capable of distinguishing both small and large quantities. Keeping those measurements is different from drawing them on a page with limited ink tones. The analogy is limited because image appearance also depends on adaptation, color, viewing conditions and spatial relationships, not only a single numerical range.

Worked example

Suppose a scene's brightest relevant luminance is 1,024 times its darkest relevant luminance. The ratio corresponds to log₂(1,024)=10 stops. If an illustrative capture system can reliably distinguish only a ratio of 256 under the chosen conditions, that is 8 stops. The scene exceeds that usable range by 2 stops, or a factor of four.
These figures are a simplified range comparison, not specifications for a real camera. The darkest usable level depends on an agreed noise or visibility criterion. Additional exposures may help record the scene, but movement, clipping and calibration affect the result. A display with different limits still requires an appropriate mapping of the captured information.

Illustrative case study

Fictional example: Willow Interiors photographs a reception area with bright outdoor windows. A single exposure either clips the view or leaves the room too noisy. The team captures a controlled bracket, checks alignment and creates an HDR representation before choosing a restrained output treatment. It reviews both the bright signage and shadow detail at delivery size. The result is evaluated for useful information and appearance, not for how dramatic an HDR effect looks.

Common misconceptions and mistakes

HDR is not synonymous with exposure merging, because broad-range capture can use other methods. It is not the same as wide color gamut, although delivery systems may support both. More bits can improve precision without automatically widening a device's physical range. Raising shadows in a clipped or noisy image does not guarantee recovered information. Tone mapping and HDR are related but distinct. Finally, an HDR label does not tell the whole story about brightness, black level, local contrast or viewing conditions.

Business use

Businesses should specify which part of the pipeline needs HDR and why. For capture, identify the scene information that must survive. For delivery, define supported formats, displays and fallback behavior. Review color management, tone mapping and output versions together. Avoid comparing camera or display claims that use different measurement conditions. Where images support analysis, retain the relevant scene-referred data and processing history rather than relying only on an attractive rendered preview. The value is preserving and presenting useful range consistently across the workflow.

People also ask

Does HDR always require several photographs?

No. Multiple exposures are one approach; sensors and other capture designs can also provide broad dynamic range.

Is an HDR image always brighter everywhere?

No. It can preserve dark regions while allowing brighter highlights. Appearance depends on content and mapping.

Can an ordinary display show an HDR file?

Software can map it to the display's capabilities, but that is not the same as reproducing the full intended HDR luminance range.

Check your understanding

A range increases from 256:1 to 1,024:1. How many stops are added?

Answer: Two stops, because the ratio increases by a factor of four, and each stop doubles the range.

Related terms

• Dynamic range
• HDR merging
• Tone mapping
• exposure
• radiance map
• bit depth
• display luminance

Learn next

Learn exposure ratios and stops, then distinguish scene range, sensor range and display range. Compare an HDR representation with its tone-mapped output before interpreting product labels or editing styles.

More in Computer vision and multimodal AI

• Hartley transform - The Hartley transform is a real-valued transform closely related to the Fourier transform.
• Hausdorff distance - Hausdorff distance measures the greatest nearest-neighbor separation between two sets of points.
• Hays - Efros scene completion - Hays - Efros scene completion is a research approach that fills missing image regions by retrieving compatible content from a very large photograph collection.
• HDR-VDP - HDR-VDP is a family of visual-difference prediction methods designed to estimate how people perceive differences between images across a broad luminance range.
• Healing brush - A healing brush is an image-retouching tool that replaces unwanted local detail while adapting the replacement to the surrounding appearance.
• Hierarchical hidden Markov model - A hierarchical hidden Markov model, or HHMM, is a probabilistic sequence model whose hidden states are organized into nested levels.

Sources

Debevec and Malik, Recovering High Dynamic Range Radiance Maps from Photographs, authors' project: https://www.pauldebevec.com/Research/HDR Primary paper: https://www.pauldebevec.com/Research/HDR/debevec-siggraph97.pdf Primary research excerpts consulted; full capture/display-specific and editorial review remain pending. Stop calculations and the fictional case are original.

Previous: HBJ model Next: HDR merging