Lossy Compression of High Dynamic Range Images and Video

1
Lossy Compression ofHigh Dynamic Range Images
and Video

• Rafal Mantiuk, Karol Myszkowski, Hans-Peter
  Seidel
• MPI Informatik, Saarbrücken, Germany

2
High Dynamic Range
luminance range log cd/m2
conventional display
human vision
3
High Dynamic Range
luminance range log cd/m2
conventional display
human vision
HDR display
4
LDR vs. HDR Images and Video
5
Why HDR Images and Video?

• Image format should be as accurate as the HVS
• Full range of visible luminance levels
• Full color gamut
• To support future displays
• Device independent

DR37-P BrightSide Technologies
6
HDR image formats

• Full HDR formats
• Radiances RGBE, logLuv TIFF, OpenEXR, JPEG HDR
• Extended dynamic range formats
• Digital Pixture Exchange (DPX)
• DICOM (medical images)
• Formats capable of HDR
• JPEG-2000, MPEG-4, TIFF

7
Color Space for HDR pixels

• Full visible range of luminance and color gamut
• Perceptually uniform
• Only positive integer numbers
• Quantization below the detection threshold
• Minimum correlation between color channels
• Direct transformation color space lt-gt XYZ

8
Luminance and Luma
luma
luminance
9
Luminance lt-gt Luma derivation
Maximum quantization error lt the detection
threshold
Adaptation to a single pixel
Inequality to equality
Solution
10
Weber Law

• Weber Law does not fit to the experimental data
  for luminance!

Weber-Fechner Law
11
Contrast Detection Models

• Threshold versus intensity function (t.v.i.)
• Contrast Sensitivity Function
• Models
• Ferwerdas t.v.i. (computer graphics)
• Blackwells t.v.i. measurements (CIE 19/2.1
  standard)
• Bartens CSF
• VDPs CSF
• CSF is a function of spatial frequency
• Conservative assumption we always take the
  maximum sensitivity

12
Contrast Detection Models t.v.i.
13
Luminance to Luma Mapping
14
Approximate Nonlinearity
15
Response of Film Negative
16
Chrominance and Chroma

• CIE 1976 Uniform Chromacity Scales (uv)

• Perceptually uniform
• 8 bits roughly sufficient precision

17
Validation of HDR Color Space

• Image Format
• A lossy compression method for OpenEXR
• OpenEXR is a popular HDR image format supported
  by Industrial Light and Magic
• Based on JPEG (DCT), but 12 bits for luma
• Very fast encoding, decoding
• Video Format
• An extension to MPEG-4 (ISO/IEC 14496-2)
• 11 bit for luma, 2x8 bit for chroma
• Demo!

18
Conclusion

• Color space for HDR images and video
• 12 bit luma, 2 x 8 bit chroma
• Derived from contrast detection data
• Meets all requirements full color gamut and
  luminance range perceptually uniform no
  countouring positive integer numbers channels
  decorrelated conversion formulas from/to XYZ
• Straightforward extension for the existing
  formats
• Proof of concept extension of MPEG and JPEG to
  HDR

19
Acknowledgements

• HDR movie sequences captured with the IMS Chips
  HDRC video camera
• See the booth
• Fire Breather video courtesy of IMS Chips
• HDR Video shooting and calibration Grzegorz
  Krawczyk

20
Thank you!