Photographic tone reproduction for digital images
Summary (2 min read)
1 Introduction
- The range of light the authors experience in the real world is vast, spanning approximately ten orders of absolute range from star-lit scenes to sun-lit snow, and over four orders of dynamic range from shadows to highlights in a single scene.
- Most of this work has used an explicit perceptual model to control the operator [Upstill 1985; Tumblin and Rushmeier 1993; Ward 1994; Ferwerda et al.
- First, current models often introduce artifacts such as ringing or visible clamping (see Section 4).
- This has led us to develop a tone reproduction technique designed for a wide variety of images, including those having a very high dynamic range .
2 Background
- The tone reproduction problem was first defined by photographers.
- Copyrights for components of this work owned by others than ACM must be honored.
- Before discussing how the Zone System is applied, the authors first summarize some relevant terminology.
- Because zones relate logarithmically to scene luminances, dynamic range can be expressed as the difference between highest and lowest distinguishable scene zones .
- For regions where loss of detail is objectionable, the photographer can resort to dodging-andburning which will locally change the development process.
3 Algorithm
- The Zone System summarized in the last section is used to develop a new tone mapping algorithm for digital images, such as those created by rendering algorithms (e.g., [Ward Larson and Shakespeare 1998]) or captured using high dynamic range photography [Debevec and Malik 1997].
- The authors are not trying to closely mimic the actual photographic process [Geigel and Musgrave 1997], but instead use the basic conceptual framework of the Zone System to manage choices in tone reproduction.
- Then, if necessary, the authors apply automatic dodging-and-burning to accomplish dynamic range compression.
3.1 Initial luminance mapping
- The authors first show how to set the tonal range of the output image based on the scene’s key value.
- Like many tone reproduction methods [Tumblin and Rushmeier 1993; Ward 1994; Holm 1996], the authors view the log-average luminance as a useful approximation to the key of the scene.
- The denominator causes a graceful blend between these two scalings.
- As mentioned in the previous section, this is not always desirable.
- For many high dynamic range images, the compression provided by this technique appears to be sufficient to preserve detail in low contrast areas, while compressing high luminances to a displayable range.
3.2 Automatic dodging-and-burning
- In traditional dodging-and-burning, all portions of the print potentially receive a different exposure time from the negative, bringing “up” selected dark regions or bringing “down” selected light regions to avoid loss of detail [Adams 1983].
- The authors choice of center-surround ratio is 1.6, which results in a difference of Gaussians model that closely resembles a Laplacian of Gaussian filter [Marr 1982].
- To choose the largest neighborhood around a pixel with fairly even luminances, the authors threshold V to select the corresponding scale sm.
- In either case the pixel’s contrast relative to the surrounding area is increased.
- In summary, by automatically selecting an appropriate neighborhood for each pixel the authors effectively implement a pixel-by-pixel dodging and burning technique as applied in photography [Adams 1983].
4 Results
- The convolutions of Equation 5 were computed using a Fast Fourier Transform (FFT).
- Ward’s contrast scale factor A global multiplier is used that aims to maintain visibility thresholds [Ward 1994].
- In Figure 11 eight different tone mapping operators are shown side by side using the Cornell box high dynamic range image as input.
- The model is slightly different from the original Cornell box because the authors have placed a smaller light source underneath the ceiling of the box so that the ceiling receives a large quantity of direct illumination, a characteristic of many architectural environments.
- The authors have also experimented with a fast approximation of the Gaussian convolution using a multiscale spline based approach [Burt and Adelson 1983], which was first used in the context of tone reproduction by [Tumblin et al. 1999], and have found that the computation is about 3.7 times faster than their Fourier domain implementation.
5 Summary
- The authors have developed a relatively simple and fast tone reproduction algorithm for digital images that borrows from 150 years of photographic experience.
- It is designed to follow their practices and is thus well-suited for applications where creating subjectively satisfactory and essentially artifact-free images is the desired goal.
Did you find this useful? Give us your feedback
Citations
1,998 citations
1,381 citations
1,226 citations
1,072 citations
793 citations
References
3,480 citations
2,615 citations
1,370 citations
1,246 citations
723 citations
Related Papers (5)
Frequently Asked Questions (13)
Q2. how many orders of absolute dynamic range can the authors reproduce on their print and screen display devices?
the range of light the authors can reproduce on their print and screen display devices spans at best about two orders of absolute dynamic range.
Q3. Why is the local operator insensitive to edge artifacts?
Because of the normalization by V1, their method is insensitive to edge artifacts normally associated with the computation of an FFT.
Q4. How many pixels wide is the Gaussian profile?
Their new local operator uses Gaussian profiles s at 8 discrete scales increasing with a factor of 1.6 from 1 pixel wide to 43 pixels wide.
Q5. What is the purpose of Equation 7?
Equation 7 is computed for the sole purpose of establishing a measure of locality for each pixel, which amounts to finding a scale sm of appropriate size.
Q6. How long does the global operator take to perform?
The total time for a 5122 image is 1.31 seconds for the local operator, which is close to interactive, while their global operator (Equation 3) performs at a rate of 20 frames per second, which the authors consider real-time.
Q7. How did the authors obtain the luminance values from the input R, G and B triplets?
The authors implemented their algorithm in C++ and obtained the luminance values from the input R, G and B triplets with L = 0.27R + 0.67G + 0.06B.
Q8. How do the authors map the luminance of a scene?
If the scene has normal-key the authors would like to map this to middle-grey of the displayed image, or 0.18 on a scale from zero to one.
Q9. What is the problem with the area around the sun in the rendering of the landscape?
the area around the sun in the rendering of the landscape is problematic for any method that attempts to bring the maximum scene luminance within a displayable range without clamping.
Q10. What was the first attempt to bridge the gap between artistic and technical aspects of photography?
Ansel Adams attempted to bridge this gap with an approach he called the Zone System [Adams 1980; Adams 1981; Adams 1983] which was first developed in the 1940s and later popularized by Minor White [White et al. 1984].
Q11. What is the difference between the highest and lowest scene zones?
Because zones relate logarithmically to scene luminances, dynamic range can be expressed as the difference between highest and lowest distinguishable scene zones (Figure 4).
Q12. How many standard deviations overlap with 1 pixel?
For practical purposes the authors would like the Gaussian profile at the smallest scale to have 2 standard deviations overlap with 1 pixel.
Q13. What is the difference between the local FFT based approximation and the global operator?
As such, the local FFT based implementation, the local spline based approximation and the global operator provide a useful trade-off between performance and quality, allowing any user to select the best operator given a specified maximum run-time.