Digital High-Dynamic Range Photography

Nikolas Coukouma

What does "Dynamic Range" mean?

• How big a range, of brightness and color, can be represented
• It's a term that applies to graphics in general
  • Recording/Capture
  • Storage
  • Editing
  • Display
• It's also used in audio
• Related measurements
  • Contrast ratio
  • Bit-depth
  • (F-)Stops

Ballpark figures

• Optimal contrast ratio

  Low

  300:1

  High

  50,000+:1

• Realistic contrast ratio

  Low

  10:1

  High

  100:1

• Bit depth

  Low

  8 bit integers per color, per pixel

  High

  32 bit floating-point per color, per pixel

Why do you want high dynamic range?

• Real life doesn't always accomodate low-dynamic range
• Humans eyes support high dynamic range
• More likely to get a get a usable image from one shot
• Bigger numbers are cooler!

HDR is not new

• Analog
  • In 1915, Prokudin Gorskii used three black and white shots, with filters, to increase range in color images
  • Dodging and burning to cope with paper prints having less range than original negatives. See W. Eugene Smith's "A Man of Mercy" from 1954
• Digital
  • Jourlin and Pinoli's work in 1988. Mostly processing, but high luminence values were stored
  • RGBE format was created in 1989 for the Radiance lighting simulation system
  • Tone mappings have been done since the 1990s. See Tumblin and Rushmeier 1993, Ward 1994, Larson et. al. 1997
  • Improving dynamic range using multiple exposures dates back to Mann 1993

Why is there more interest now?

• Digital devices have finally caught up
• Cameras
  • Are small
  • Have a lot of pixels - 6 to 13 million!
  • Can capture almost as much light as film
• Hard drives and flash memory have gotten bigger
  • 512MB card costs $30 and holds 150 6MP JPEG images
  • Same card holds about 35 HDR images
• Displays
  • Went down in size with LCDs
  • Also went down in image quality
  • Have gotten big again (19" common, 30" not unheard of)

HDR now available

• Cameras
  • Fuji's "Super CCD SR" (used in their S3 Pro)
• Storage
  • OpenEXR
  • TIFF LogLuv
  • Floating point TIFF
  • RGBE
  • Portable Float Maps
• Editing
  • HDRShop
  • Idruna Photogenics
• Display
  • BrightSide DR37-P

But what can I do today?

• HDR input and output are rare and expensive
• So, let's work with what we have
  • Exposure blending
  • Recovering HDR from multiple LDR
  • Mapping HDR into LDR
  • Backwards compatible format - JPEG/JFIF

Exposure blending

• At no point is there an HDR image
• Just blend parts of multiple LDR exposures
• Usually done by hand
• Simple in concept, but hard to do well

Recovering an HDR image

• The general process:
  1. Take multiple LDR photos, at different shutter speeds
  2. Figure out and adjust for problems with the images
     • Images not aligned
     • Cameras aren't perfect
  3. Combine them into one image, weighting by exposure

Imperfect cameras

• Cameras don't record extremely bright or dark images well
• Should be a nice power/gamma curve
• To adjust, apply a weighting function

Pentax *ist DS response

Multiple LDR images

1/3	1/8	1/20
1/60	1/180	1/500

Luma frequencies

1/3	1/8	1/20
1/60	1/180	1/500

The resulting HDR image

• room.exr - OpenEXR format
• View using exrdisplay, Photosphere, pfsview
• This one was actually made from 18 exposures, from 1/750 to 1/2
• 6 exposures is more common than 18, but it depends on the scene

Mapping HDR into LDR

• Basic goal: make an image that 'fits' in an LDR format/display
• Aim to make it natural, as if you were there
• Called tone mapping
• Algorithms are called operators
• Many tone mapping operators exist with different motivations

Characteristics to consider

• Brightness
• Contrast
• Color reproduction
• Reproduction of details
• Simulation of glare
• Reproduction of detail
• Visual acuity
• Artifacts

Middle exposure

Slice from HDR

Drago mapping

Fattal mapping

Side by side

Middle exposure	Slice from HDR
Drago mapping	Fattal mapping

Side by side: luma

Middle exposure	Slice from HDR
Drago mapping	Fattal mapping

Backwards compatible JPEG/JFIF

1. Apply a tone mapping to create regular 8-bit part
   • Must maintain hue
2. Apply normal JPEG compression to tone mapped image
3. Calculate the ratio image
   • Divide luma of HDR by luma of LDR
4. Wavelet compress ratio image
5. Store in application data markers
   • 16 available, 2 already reserved
   • Each limited to 64KB
   • 14 * 64KB = 896KB

Details get messy

Meta

• Download this presentation - all images and source code
• Utilities used:
  • pfstools - general HDR utils
  • pfscalibration - calibrate cameras, recover HDR images
  • pfstmo - tone mapping operators
  • Gnuplot - plotting data
• Some links
• HDR JPEG/JFIF diagram is from the paper

That's all!

Questions?