CS Map

HDR

Also known as: High Dynamic Range, HDR10, Dolby Vision, HDR display, tone mapping

supplemental intermediate concept 5 min read · Updated 2026-06-17

High Dynamic Range imaging captures and displays a wider range of luminance — from very dark to very bright — than standard displays, requiring new formats, tone mapping, and colour science to render correctly.

Primary domain
Graphics & Media
Sub-category
Photograph Manipulation & Image Compression

In simple terms

A candle flame and a snow-covered landscape on a sunny day differ in brightness by about a factor of one million. A standard (SDR) display can only show a factor of about 1,000. High Dynamic Range (HDR) displays show up to a factor of one million — a much closer approximation of what eyes see in reality. A candle in a dark room actually looks like a point of fire rather than a washed-out blob. HDR requires new display hardware (local-dimming LCD or OLED), new content formats (HDR10, Dolby Vision), and new software to correctly map the wide range of brightness values to what each display can show.

The Visual Map

graph TD
    scene["Scene luminance\n0 – 10,000+ nits"] --> enc["Encode: 10/12-bit + PQ or HLG\n(HDR10 · HDR10+ · Dolby Vision · HLG)"]
    enc --> tone["Tone mapping\ncompress to display's peak nits\n(Reinhard · ACES · filmic)"]
    tone --> disp["Display\nOLED ~0 nit black, 1000–2000 peak"]
    sdr["SDR baseline\n~100–300 nits, 8-bit, gamma 2.2"] -.contrast.-> disp
    enc --> meta["Metadata:\nstatic (HDR10) vs dynamic (DV/HDR10+)"]

More detail

Dynamic range is the ratio of brightest to darkest luminance a display can show simultaneously, measured in nits (cd/m²). SDR is typically 100–300 nits peak with ~1000:1 contrast; HDR reaches 600–10,000+ nits peak (OLED ~1000–2000 with near-0 black, giving ~1,000,000:1 contrast). HDR also needs more bit depth — 10-bit (1024 levels) or 12-bit instead of SDR’s 8-bit — to span that range without visible banding.

Standards include HDR10 (open, royalty-free, PQ transfer function, 10-bit, static metadata describing the whole programme), HDR10+ (Samsung’s open standard with dynamic per-scene metadata), Dolby Vision (proprietary, 12-bit, dynamic per-frame metadata, licensed displays), and HLG (broadcast-oriented, backward-compatible with SDR, no metadata). Transfer functions differ from SDR’s gamma 2.2: HDR uses PQ (SMPTE ST 2084), a perceptually-uniform curve with absolute luminance values, or HLG’s log-gamma hybrid that degrades gracefully to SDR.

Tone mapping is the crucial step: when 10,000-nit content is shown on a 1000-nit TV, the brightness must be compressed — algorithms like ACES, Reinhard, and filmic preserve highlight and shadow detail differently. Games render into a floating-point (FP16) buffer internally and map to the display’s PQ signal on output. The recurring problems are that content mastered on a 4000-nit reference monitor looks different on a 600-nit consumer panel (dynamic metadata helps), and that the same “HDR” label spans wildly different hardware capabilities.

Under the Hood

Tone mapping is where HDR meets the limits of real hardware. The Reinhard operator — L / (1 + L) — is the classic compressor: it leaves dark detail almost untouched while gently rolling very bright values toward the display’s peak, so nothing simply clips to white:

def reinhard(L):                 # L = scene luminance (relative)
    return L / (1 + L)           # maps [0, ∞) → [0, 1)

def reinhard_extended(L, Lwhite):    # lets values up to Lwhite reach full white
    return L * (1 + L / (Lwhite * Lwhite)) / (1 + L)

print(f"{'scene L':>8}{'Reinhard':>12}{'extended(8)':>14}")
for L in [0.1, 0.5, 1, 2, 4, 8, 20, 100]:
    print(f"{L:>8.1f}{reinhard(L):>12.3f}{reinhard_extended(L, 8):>14.3f}")

Notice dark values (0.1 → ~0.09) pass through nearly linearly while bright values (100 → ~0.99) are squeezed into the top of the range — preserving shadow detail and highlight separation instead of blowing out.

Engineering Trade-offs

  • Peak brightness vs panel cost/power. Higher nits and local dimming make highlights pop but cost panel complexity, power, and heat; cheap “HDR” panels accept a signal they can’t actually display.
  • Static vs dynamic metadata. HDR10’s single metadata set is simple and universal but tone-maps every scene the same; Dolby Vision/HDR10+ adapt per scene at the cost of licensing or limited support.
  • Bit depth vs bandwidth. 10–12 bit removes banding across the wide range but increases data rate and pipeline width versus 8-bit SDR.
  • Authoring control vs consistency. Mastering for a bright reference monitor maximises impact but looks different across the fragmented field of consumer displays.

Real-world examples

  • Netflix serves Dolby Vision to capable TVs with HDR10 as the fallback.
  • Apple’s Extended Dynamic Range (EDR) on macOS and the Pro Display XDR support up to ~1600 nits sustained for HDR reference work.
  • Xbox Series X and PlayStation 5 support HDR10 and Dolby Vision; most modern games ship HDR rendering paths.
  • iPhone Pro models record video in HDR (HLG/Dolby Vision) and have ~2000-nit displays.

Common misconceptions

  • “Any HDR badge means better quality.” Many cheap “HDR” monitors lack local dimming or sufficient peak brightness; they accept an HDR signal but can’t render the wide range. Look for DisplayHDR 1000+ or OLED.
  • “HDR is only for movies.” Game HDR, professional UI (Apple’s EDR), and medical imaging all use HDR for different reasons.

Try it yourself

Apply Reinhard tone mapping and watch huge scene luminances fold into a displayable 0–1 range without clipping (python3 only):

python3 - <<'EOF'
reinhard = lambda L: L/(1+L)
print(f"{'scene nits (rel)':>16}{'displayed (0-1)':>16}")
for L in [0.05, 0.2, 1, 3, 10, 50, 500]:
    print(f"{L:>16.2f}{reinhard(L):>16.3f}")
EOF

Learn next

  • Color management — HDR needs wide color spaces (Rec. 2020, P3) and accurate conversions
  • Color space — the primaries and transfer functions (PQ/HLG) HDR depends on
  • Anti-aliasing — another facet of display image quality
  • Subpixel rendering — a related display-precision technique for text

Neighborhood

A visual companion to the relationships above. Click any node to visit that topic.