High Dynamic Range

1
High Dynamic Range

• Imaging and Video Coding

By Danny Luong
2
Topics Covered

• HDR Imaging Overview
• HDR Image Generation/Examples
• HDR Image Compression
• Tone Mapping
• HDR Video Overview
• HDR Video Generation
• HDR Video Compression

3
Motivation for High Dynamic Range

• A standard dynamic range image encodes luminance
  values with 8 bits allowing only a range of
  values from 0 to 255.
• This range limits the ability of the image to
  represent the true-to-life luminance and
  intensity levels of a given scene.
• The dynamic range ratio humans perceive is
  approximately 100001, while standard cameras are
  only capable of capturing 4001.

4
What is High Dynamic Range?

• High dynamic range imaging addresses this issue
  of limited intensity ranges by using more bits to
  represent luminance values (up to 32 bits).
• HDR images can be generated using a set of low
  dynamic range images, or using specialized
  equipment.
• There are issues with viewing HDR images on
  standard display devices (tone mapping).

5
HDR Image Example

• LDR Image

• HDR Image

6
Pros, Cons, Applications

• Increased scene detail is preserved when
  compared to LDR images.
• Larger range of luminance values can be
  represented.
• Luminance values can be floating point, rather
  than integer values.
• - Larger file sizes
• Application More realistic motion blur

7
Motion blur of LDR/HDR images

• Original Photo

• Simulated blur-LDR

Simulated Blur-HDR
Real motion blur-HDR
8
Radiance Maps

• Radiance maps are used in generating a HDR image
  from a set of LDR images.
• Radiance maps have pixel values that are
  proportional to the true radiance values in a
  scene.
• How? Given a set of LDR images with different
  exposures, use an algorithm to combine the
  luminance values. (There are a number of existing
  algorithms)

9
Radiance Map

• HDR Image

• False Color Radiance map

10
HDR from LDR images

• Six LDR images taken with varying exposure values

• HDR image generated from six LDR images

11
Radiance Maps Generation

• Equation for irradiance of a pixel, i
• Radiance map using set of pictures
• Weighting function

• Camera response characteristic curve
• Method by Debevc Malik, 1997

12
HDR Image Compression

• The need for compression is apparent due to
  increased file size of HDR content.
• Different lossy and lossless compression schemes
  for HDR images exist, including run length
  coding, wavelet compression, subband encoding.
• OpenEXR (near-lossless wavelet) has 21
  compression ratio, while lossy JPEG2000 at rate
  0.05 is indistinguishable from original.

13
HDR Image Compression II

• Standard JPEG2000 block diagram
• HDR JPEG2000 compression extension

14
HDR Image Compression III

• Map HDR pixel values to log domain and convert
  floating point values to integers
• where

• Use JPEG2000 encoder with RGB to YCbCr space.
• Use wavelet transform and quantize with stepsize

15
Tone Mapping

• Tone mapping is a method of representing a space
  of values with a large range into a space with
  limited range.
• Tone mapping is needed because current display
  devices cannot physically reproduce luminance
  values apparent in the real world.
• There are many different methods of tone mapping
  (local, non-local).
• Tone map for display on LCD, projector, etc.

16
HDR Video

• Like HDR imaging, there are different possible
  ways to obtain HDR video.
• Methods include HDR capable capture video cameras
  (costly), fixed mask, adaptive light modulator,
  and varying exposures for alternate frames.
• Like HDR imaging, display to standard monitors
  requires tone mapping, but normalization across
  frames must be applied.

17
HDR Video Example

• Top original frame sequence

• Bottom HDR frame sequence

18
Generating HDR Video

• The method we will cover is alternating frame
  exposures, also called HDR stitching.
• There is a need to create frames with differing
  exposure values to generate radiance maps at each
  instant.
• A warping process is used to generate
  interpolated frames.
• Any method of radiance map generation can be used
  on the pairs of frames.

19
HDR Stitching
Next (Short Exposure) Frame
Prev (Short Exposure) Frame
Generated Frame
Generated Frame
Generated Frame
Generated Frame
Current (Long Exposure) Frame
20
HDR Stitching II
HDR Frame
Generated Frame
Generated Frame
Radiance Map
Generated Frame
Generated Frame
Current (Long Exposure) Frame
21
HDR Video Coding

• HDR video can be compressed using MJPEG2000,
  where each individual frame can be compressed
  using the HDR JPEG2000 compression method covered
  earlier.
• Other methods exist for compression, one based on
  MPEG4 is by Mantiuk, et al. (2004)

22
HDR Video Coding II

• Steps in coding (Mantiuk, et al., 2004)
• Convert HDR XYZ color space to Lpuv
• Quantization of luminance values
• Motion estimation
• Perform DCT
• EITHER quantize OR
• Use hybrid method
• Hybrid method reduces
• edge artifacts (run length
• encoding of hi-freq content)

23
References

• http//www.mpi-inf.mpg.de/resources/hdr/
• http//en.wikipedia.org/wiki/High_dynamic_range_im
  aging
• http//en.wikipedia.org/wiki/Tone_mapping
• http//www.debevec.org/Research/HDR/
• http//www.cybergrain.com/tech/hdr/
• http//www.naturescapes.net/072006/rh0706_1.htm
• DEBEVEC, P. E., AND MALIK, J. 1997. Recovering
  high dynamic range radiance maps from
  photographs. In Proc. ACM SIGGRAPH 97, T.
  Whitted, Ed., 369378.
• KANG, S.B., ET AL. 2003. High Dynamic Range
  Video. In In ACM Trans. Graph., vol. 22, no.
  3,2003, pp. 319-325.
• MANTIUK, R., ET AL. 2004. Perception-Motivated
  High-Dynamic-Range Video Encoding. In ACM Trans.
  Graph., vol. 23, no. 3,2004, pp. 733-741.
• XU, R., ET AL, 2005. High-Dynamic-Range
  Still-Image Encoding in JPEG 2000. In IEEE
  Computer Graphics and Applications., vol. 25, no.
  6, 2005, pp.57-64.