Back

Gradient domain high dynamic range compression

Raanan FattalDani LischinskiMichael Werman

article2002SIGGRAPH1,521 citations

Presents a tone mapping approach that compresses high dynamic range images by attenuating large luminance gradients across multiple scales and solving a Poisson equation, preserving fine local contrast while preventing halo artifacts.

Listen

Real-world scenes often exhibit extreme contrasts between brightly lit and deeply shadowed areas, creating dynamic ranges that exceed 100,000:1. While modern digital sensors and multi-exposure photography can easily capture these high dynamic range scenes, standard display devices like monitors and printers can only reproduce dynamic ranges below 100:1. Existing compression techniques either wash out local contrast or introduce unnatural visual distortions, such as artificial halos and embossed edges around bright boundaries. The main objective of the article is to demonstrate an efficient and robust gradient-domain compression method that compresses drastic dynamic ranges into standard display formats while preserving fine detail and preventing visual artifacts.

The evaluated approach operates on the gradient field—the rates of local luminance change—of an image's logarithmic representation. Large gradients corresponding to drastic lighting transitions are identified across multiple scales and attenuated, while small gradients corresponding to fine textures are preserved. Because altering gradients creates a field that cannot be directly integrated back into an image, the method solves a Poisson differential equation using a standard multigrid numerical solver to reconstruct a compressed, low dynamic range image. The authors tested this algorithm on synthetic and real-world high dynamic range radiance maps, panoramic video mosaics, standard high-contrast photographs, and medical fluoroscopic scans.

The findings demonstrate that this gradient attenuation method compresses extreme dynamic ranges (such as scenes exceeding 250,000:1) into standard displays with high fidelity. Compared to previous state-of-the-art approaches, the method eliminates halo artifacts, avoids artificial edge outlines, and preserves local textures in both bright and shadowed regions. Computationally, the algorithm operates in linear time relative to pixel count, processing standard high-resolution images in roughly one to five seconds—orders of magnitude faster than multi-minute partial differential equation methods like the low curvature image simplifier. Additionally, the approach effectively enhances ordinary standard-exposure photographs and medical imagery by uncovering obscured details in dark regions without introducing noise or halos.

These results indicate that gradient-domain processing offers a practical, production-ready solution for digital photography pipelines, video processing, and imaging software. Organizations handling image visualization can achieve superior visual quality at significantly lower computational and operational costs. Stakeholders should consider adopting this framework for high dynamic range tone mapping and standard image enhancement workflows. Future efforts should focus on integrating perceptual human vision models for specialized lighting and visibility design, as well as extending gradient-field reconstruction to tasks such as image denoising and non-photorealistic rendering.

While confidence in the algorithm's performance is high across diverse test scenes, the authors note that the method is designed for visual appeal and detail preservation rather than exact psychophysical simulation of human eye adaptation. In applications where absolute photometric precision or strict human visual fidelity is required, practitioners should exercise appropriate caution.

The Hebrew University of Jerusalemhigh resolution imagewide dynamic rangehigh dynamic range compressiongradient domaingradient attenuation
Fattal et al (2002).pdf
  • Paper: Recovering high dynamic range radiance maps from photographs, Paul E. Debevec et al. (1997). This seminal work establishes the foundational technique for recovering high dynamic range radiance maps from exposure sequences, providing the input representation that gradient domain compression operates upon.
  • Paper: Poisson image editing, Patrick Pérez et al. (2003). This paper builds directly upon gradient-domain Poisson image reconstruction concepts to establish a broader framework for seamless image editing, cloning, and guided interpolation.
  • Paper: Single image dehazing, Raanan Fattal (2008). This work extends computational techniques for manipulating image luminance and contrast to the inverse problem of single-image haze removal.
  • Paper: Guided Image Filtering, Kaiming He et al. (2010). This paper provides an efficient, explicit edge-preserving alternative to optimization- and Poisson-based gradient manipulation for tasks like dynamic range compression and detail enhancement.
Cover for Gradient domain high dynamic range compression

Abstract

We present a new method for rendering high dynamic range images on conventional displays. Our method is conceptually simple, computationally efficient, robust, and easy to use. We manipulate the gradient field of the luminance image by attenuating the magnitudes of large gradients. A new, low dynamic range image is then obtained by solving a Poisson equation on the modified gradient field. Our results demonstrate that the method is capable of drastic dynamic range compression, while preserving fine details and avoiding common artifacts, such as halos, gradient reversals, or loss of local contrast. The method is also able to significantly enhance ordinary images by bringing out detail in dark regions.

Topics

high resolution imagewide dynamic range

Keywords

high dynamic range compressiongradient domaingradient attenuationtone reproduction operatorslow dynamic range imageadaptive histogram equalizationluminance gradientsgradient fieldHDR imagesimage enhancement

Table of Contents

  • 1 Introduction
  • 2 Previous work
  • 3 Gradient domain HDR compression
  • 4 Gradient attenuation function
  • 5 Implementation
  • 6 Results
  • 7 Conclusions and Future Work
  • Acknowledgments
  • References

Knowls

  1. Knowl 1 — Gradient Domain High Dynamic Range Tone Mapping Framework

    model/method

    High dynamic range (HDR) compression maps high-contrast radiance images onto low dynamic range (LDR) displays by manipulating the gradient field of the luminance image in the logarithmic domain. Given an HDR luminance map L(x,y)L(x, y), the method operates on the log-luminance H(x,y)=log⁡(L(x,y))H(x, y) = \log(L(x, y)).

    Large-scale illumination variations in H(x,y)H(x, y) manifest as large gradient magnitudes, while local texture and surface details correspond to smaller gradient magnitudes. The method computes a spatially varying attenuation factor Φ(x,y)∈(0,1]\Phi(x, y) \in (0, 1] that selectively suppresses large gradient magnitudes while leaving smaller gradient magnitudes uncompressed, producing a target vector field:

    G(x,y)=∇H(x,y)Φ(x,y)G(x, y) = \nabla H(x, y) \Phi(x, y)

    Because the modified gradient field G(x,y)G(x, y) is generally non-conservative, a compressed log-luminance image I(x,y)I(x, y) is reconstructed by solving a 2D Poisson equation that minimizes the L2L_2 difference ∥∇I−G∥2\|\nabla I - G\|_2. Exponentiating the resulting image yields the output compressed luminance map:

    Lout(x,y)=exp⁡(I(x,y))L_{\text{out}}(x, y) = \exp(I(x, y))

    which is subsequently normalized and colorized for display.

  2. Knowl 2 — Poisson Reconstruction from Non-Conservative Gradient Fields

    model/method

    When a 2D gradient field ∇H(x,y)\nabla H(x, y) is modified by a spatially varying attenuation function Φ(x,y)\Phi(x, y) to form G(x,y)=(Gx(x,y),Gy(x,y))=∇H(x,y)Φ(x,y)G(x, y) = (G_x(x, y), G_y(x, y)) = \nabla H(x, y) \Phi(x, y), the vector field G(x,y)G(x, y) is generally not conservative (that is, ∂Gx∂y≠∂Gy∂x\frac{\partial G_x}{\partial y} \neq \frac{\partial G_y}{\partial x}) and cannot be integrated directly.

    To reconstruct a potential function I(x,y)I(x, y) whose gradient best approximates G(x,y)G(x, y) in the least-squares sense, the method minimizes the objective integral:

    ∬∥∇I(x,y)−G(x,y)∥2 dx dy=∬[(∂I∂x−Gx)2+(∂I∂y−Gy)2] dx dy\iint \|\nabla I(x, y) - G(x, y)\|^2 \, dx \, dy = \iint \left[ \left(\frac{\partial I}{\partial x} - G_x\right)^2 + \left(\frac{\partial I}{\partial y} - G_y\right)^2 \right] \, dx \, dy

    Applying the Euler-Lagrange equation:

    ∂F∂I−ddx(∂F∂Ix)−ddy(∂F∂Iy)=0\frac{\partial F}{\partial I} - \frac{d}{dx}\left(\frac{\partial F}{\partial I_x}\right) - \frac{d}{dy}\left(\frac{\partial F}{\partial I_y}\right) = 0

    yields the linear Poisson equation:

    ∇2I=div G\nabla^2 I = \text{div} \, G

    where ∇2I=∂2I∂x2+∂2I∂y2\nabla^2 I = \frac{\partial^2 I}{\partial x^2} + \frac{\partial^2 I}{\partial y^2} is the Laplacian operator and div G=∂Gx∂x+∂Gy∂y\text{div} \, G = \frac{\partial G_x}{\partial x} + \frac{\partial G_y}{\partial y} is the divergence of GG. The Poisson equation is solved subject to Neumann boundary conditions ∇I⋅n=0\nabla I \cdot \mathbf{n} = 0, where n\mathbf{n} is the boundary normal. This defines I(x,y)I(x, y) up to an arbitrary additive constant, which is resolved during display mapping.

  3. Knowl 3 — Multi-Scale Gradient Attenuation Algorithm

    algorithm

    To capture luminance edges occurring at different spatial scales without producing halo artifacts, gradient attenuation factors are computed across a Gaussian pyramid and accumulated onto the finest resolution level before modifying the gradients.

    Input: Log-luminance image H0H_0 of dimensions W×HW \times H, threshold parameter α\alpha, exponent β\beta
    Output: Attenuated gradient field G=(Gx,Gy)G = (G_x, G_y) at full resolution
    Construct Gaussian pyramid H0,H1,…,HdH_0, H_1, \dots, H_d such that min⁡(width(Hd),height(Hd))≥32\min(\text{width}(H_d), \text{height}(H_d)) \ge 32
    for each level kk from 00 to dd:
        Compute gradient ∇Hk(x,y)=(Hk(x+1,y)−Hk(x−1,y)2k+1,Hk(x,y+1)−Hk(x,y−1)2k+1)\nabla H_k(x, y) = \left(\frac{H_k(x+1, y) - H_k(x-1, y)}{2^{k+1}}, \frac{H_k(x, y+1) - H_k(x, y-1)}{2^{k+1}}\right)
        Compute scaling factor ϕk(x,y)=α∥∇Hk(x,y)∥(∥∇Hk(x,y)∥α)β\phi_k(x, y) = \frac{\alpha}{\|\nabla H_k(x, y)\|} \left(\frac{\|\nabla H_k(x, y)\|}{\alpha}\right)^\beta
    Set accumulated attenuation Φd(x,y)=ϕd(x,y)\Phi_d(x, y) = \phi_d(x, y)
    for level kk from d−1d-1 down to 00:
        Upsample Φk+1\Phi_{k+1} to the resolution of level kk using bilinear interpolation L\mathcal{L}
        Set Φk(x,y)=L(Φk+1)(x,y)⋅ϕk(x,y)\Phi_k(x, y) = \mathcal{L}(\Phi_{k+1})(x, y) \cdot \phi_k(x, y)
    Set full-resolution attenuation Φ(x,y)=Φ0(x,y)\Phi(x, y) = \Phi_0(x, y)
    Compute modified gradient field G(x,y)=∇H0(x,y)Φ(x,y)G(x, y) = \nabla H_0(x, y) \Phi(x, y)
    return GG

    Because gradient attenuation factors Φ(x,y)\Phi(x, y) are multiplied across all pyramid scales and applied exclusively at the base resolution H0H_0, all gradient modifications occur at a single resolution level, preventing the halo artifacts typical of independent multi-band compression.

  4. Knowl 4 — Gradient Attenuation Scaling Function

    equation

    At each pyramid level kk, the local gradient attenuation scaling factor ϕk(x,y)\phi_k(x, y) is defined as a two-parameter power-law function of the gradient magnitude ∥∇Hk(x,y)∥\|\nabla H_k(x, y)\|:

    ϕk(x,y)=α∥∇Hk(x,y)∥(∥∇Hk(x,y)∥α)β=(∥∇Hk(x,y)∥α)β−1\phi_k(x, y) = \frac{\alpha}{\|\nabla H_k(x, y)\|} \left( \frac{\|\nabla H_k(x, y)\|}{\alpha} \right)^\beta = \left( \frac{\|\nabla H_k(x, y)\|}{\alpha} \right)^{\beta - 1}

    where:

    • α>0\alpha > 0 is a reference gradient threshold. Gradients with magnitude ∥∇Hk(x,y)∥=α\|\nabla H_k(x, y)\| = \alpha receive a scaling factor of 11. In practice, α\alpha is set to 0.10.1 times the average gradient magnitude across the image.
    • β∈(0,1)\beta \in (0, 1) is the compression exponent. When β<1\beta < 1, gradient magnitudes exceeding α\alpha are attenuated (ϕk<1\phi_k < 1), with stronger attenuation applied to progressively larger gradients, while gradient magnitudes below α\alpha are slightly amplified. In practice, β\beta is set between 0.80.8 and 0.90.9.

  5. Knowl 5 — Finite Difference Discretization and Multigrid Solver for Gradient Integration

    algorithm

    The Poisson equation ∇2I=div G\nabla^2 I = \text{div} \, G with Neumann boundary conditions ∇I⋅n=0\nabla I \cdot \mathbf{n} = 0 is discretized on a uniform pixel grid with unit spacing (h=1h = 1).

    The discrete Laplacian ∇2I(x,y)\nabla^2 I(x, y) uses the standard 5-point stencil:

    ∇2I(x,y)≈I(x+1,y)+I(x−1,y)+I(x,y+1)+I(x,y−1)−4I(x,y)\nabla^2 I(x, y) \approx I(x+1, y) + I(x-1, y) + I(x, y+1) + I(x, y-1) - 4 I(x, y)

    The input gradient ∇H(x,y)\nabla H(x, y) is computed using forward differences:

    ∇H(x,y)≈(H(x+1,y)−H(x,y), H(x,y+1)−H(x,y))\nabla H(x, y) \approx (H(x+1, y) - H(x, y), \, H(x, y+1) - H(x, y))

    The divergence div G(x,y)\text{div} \, G(x, y) of the attenuated gradient field G=(Gx,Gy)G = (G_x, G_y) is computed using backward differences:

    div G(x,y)≈Gx(x,y)−Gx(x−1,y)+Gy(x,y)−Gy(x,y−1)\text{div} \, G(x, y) \approx G_x(x, y) - G_x(x-1, y) + G_y(x, y) - G_y(x, y-1)

    Combining forward differences for the gradient with backward differences for the divergence matches the central difference discretization of the Laplacian.

    At the domain boundary, zero-derivative conditions are enforced (for example, I(−1,y)−I(0,y)=0I(-1, y) - I(0, y) = 0 on the left edge). The resulting sparse linear system has 5 non-zero entries per row and is solved to convergence in O(n)O(n) operations for an nn-pixel image using the Full Multigrid Algorithm with Gauss-Seidel relaxation sweeps.

  6. Knowl 6 — Color Reconstruction Formula with Saturation Control

    equation

    After reconstructing the compressed log-luminance image I(x,y)I(x, y) and exponentiating to obtain the display luminance Lout(x,y)=exp⁡(I(x,y))L_{\text{out}}(x, y) = \exp(I(x, y)), color values for each pixel channel Cout∈{Rout,Gout,Bout}C_{\text{out}} \in \{R_{\text{out}}, G_{\text{out}}, B_{\text{out}}\} are computed from the original input color channels Cin∈{Rin,Gin,Bin}C_{\text{in}} \in \{R_{\text{in}}, G_{\text{in}}, B_{\text{in}}\} and original luminance Lin(x,y)L_{\text{in}}(x, y) according to:

    Cout(x,y)=(Cin(x,y)Lin(x,y))sLout(x,y)C_{\text{out}}(x, y) = \left( \frac{C_{\text{in}}(x, y)}{L_{\text{in}}(x, y)} \right)^s L_{\text{out}}(x, y)

    The exponent parameter ss controls the color saturation of the resulting compressed image, with practical values typically chosen in the range s∈[0.4,0.6]s \in [0.4, 0.6] to avoid unnatural oversaturation.

  7. Knowl 7 — Tone Mapping Performance and Halo Artifact Elimination

    empirical result

    Gradient domain HDR compression successfully compresses high dynamic range radiance maps exceeding ratios of 25,000:125,000:1, 100,000:1100,000:1, and 250,000:1250,000:1 into standard display ranges while avoiding halo artifacts, gradient reversals, and over-sharpening outlines.

    On an 1800 MHz Pentium 4 processor, the multigrid Poisson solver achieves the following execution times:

    • A 512×384512 \times 384 HDR image is compressed in 1.11.1 seconds.
    • A 1024×7681024 \times 768 HDR image is compressed in 4.54.5 seconds.
    • A 751×1130751 \times 1130 HDR image is compressed in 5.05.0 seconds.

    In comparison, the Low Curvature Image Simplifier (LCIS) method requires approximately 8.58.5 minutes (510510 seconds) on the same 751×1130751 \times 1130 image, introduces embossed/bumpy outline artifacts around high-contrast edges, and requires tuning 8 parameters, whereas the gradient domain method requires tuning only two primary parameters (α\alpha and β\beta). Global tone reproduction curves (such as histogram equalization variants) wash out local contrasts and obscure dark regions in scenes with widely distributed luminances.

  8. Knowl 8 — Contrast Enhancement of Low Dynamic Range and Fluoroscopic Images

    empirical result

    The gradient domain compression pipeline enhances ordinary 8-bit low dynamic range (LDR) photographs and medical images without requiring HDR input. Applying gradient attenuation (β∈[0.8,0.9]\beta \in [0.8, 0.9]) to the logarithm of an LDR image, solving the Poisson equation, and linearly mapping the reconstructed signal back to the [0,255][0, 255] display range amplifies subtle local contrast in deeply shadowed or low-contrast regions.

    Compared to standard enhancement methods:

    • Global gamma correction and standard histogram equalization wash out bright areas or fail to recover shadow textures.
    • Contrast-limited adaptive histogram equalization (CLAHE) recovers local details but introduces visible halo artifacts around high-contrast boundaries (such as rooflines against the sky).
    • Gradient domain compression reveals shadowed foreground textures (such as brick walls and foliage) and hidden anatomical structures in low-contrast fluoroscopic femur X-rays (such as the femur canal) without introducing boundary halos.

Coverage note — No substantial contributed material was omitted.

References

  1. 1.AGGARWAL, M., AND AHUJA, N. 2001. High dynamic range panoramic imaging. In Proc. IEEE ICCV, vol. I, 2–9.
  2. 2.AGGARWAL, M., AND AHUJA, N. 2001. Split aperture imaging for high dynamic range. In Proc. IEEE ICCV, vol. II, 10–17.
  3. 3.CHIU, K., HERF, M., SHIRLEY, P., SWAMY, S., WANG, C., AND ZIMMERMAN, K. 1993. Spatially nonuniform scaling functions for high contrast images. In Proc. Graphics Interface ’93, Morgan Kaufmann, 245–253.
  4. 4.COHEN, J., TCHOU, C., HAWKINS, T., AND DEBEVEC, P. 2001. Real-time high-dynamic range texture mapping. In Rendering Techniques 2001, S. J. Gortler and K. Myszkowski, Eds. Springer-Verlag, 313–320.
  5. 5.DEBEVEC, P. E., AND MALIK, J. 1997. Recovering high dynamic range radiance maps from photographs. In Proc. ACM SIGGRAPH 97, T. Whitted, Ed., 369–378.
  6. 6.DEBEVEC, P. 1998. Rendering synthetic objects into real scenes: Bridging traditional and image-based graphics with global illumination and high dynamic range photography. In Proc. ACM SIGGRAPH 98, M. Cohen, Ed., 189–198.
  7. 7.DICARLO, J. M., AND WANDELL, B. A. 2001. Rendering high dynamic range images. In Proceedings of the SPIE: Image Sensors, vol. 3965, 392–401.
  8. 8.FRANKOT, R. T., AND CHELLAPPA, R. 1988. A method for enforcing integrability in shape from shading algorithms. IEEE Transactions on Pattern Analysis and Machine Intelligence 10, 4 (July), 439–451.
  9. 9.HARRIS, J. W., AND STOCKER, H. 1998. Handbook of Mathematics and Computational Science. Springer-Verlag.
  10. 10.HORN, B. K. P. 1974. Determining lightness from an image. Computer Graphics and Image Processing 3, 1 (Dec.), 277–299.
  11. 11.JOBSON, D. J., RAHMAN, Z., AND WOODELL, G. A. 1997. A multi-scale Retinex for bridging the gap between color images and the human observation of scenes. IEEE Transactions on Image Processing 6, 7 (July), 965–976.
  12. 12.LAND, E. H., AND MCCANN, J. J. 1971. Lightness and Retinex theory. Journal of the Optical Society of America 61, 1 (Jan.), 1–11.
  13. 13.NAYAR, S. K., AND MITSUNAGA, T. 2000. High dynamic range imaging: Spatially varying pixel exposures. In Proc. IEEE CVPR.
  14. 14.PATTANAIK, S. N., FERWERDA, J. A., FAIRCHILD, M. D., AND GREENBERG, D. P. 1998. A multiscale model of adaptation and spatial vision for realistic image display. In Proc. ACM SIGGRAPH 98, M. Cohen, Ed., 287–298.
  15. 15.PERONA, P., AND MALIK, J. 1990. Scale-space and edge detection using anisotropic diffusion. IEEE Transactions on Pattern Analysis and Machine Intelligence 12, 7 (July), 629–639.
  16. 16.PIZER, S. M., AMBURN, E. P., AUSTIN, J. D., CROMARTIE, R., GESELOWITZ, A., GREER, T., TER HAAR ROMENY, B., ZIMMERMAN, J. B., AND ZUIDERVELD, K. 1987. Adaptive histogram equalization and its variations. Computer Vision, Graphics, and Image Processing 39, 3 (Sept.), 355–368.
  17. 17.PRESS, W. H., TEUKOLSKY, S. A., VETTERLING, W. T., AND FLANNERY, B. P. 1992. Numerical Recipes in C: The Art of Scientific Computing, 2nd ed. Cambridge University Press.
  18. 18.RAMAMOORTHI, R., AND HANRAHAN, P. 2001. A signal-processing framework for inverse rendering. In Proc. ACM SIGGRAPH 2001, E. Fiume, Ed., 117–128.
  19. 19.SCHECHNER, Y. Y., AND NAYAR, S. K. 2001. Generalized mosaicing. In Proc. IEEE ICCV, vol. I, 17–24.
  20. 20.SCHLICK, C. 1994. Quantization techniques for visualization of high dynamic range pictures. In Photorealistic Rendering Techniques, Springer-Verlag, P. Shirley, G. Sakas, and S. Müller, Eds., 7–20.
  21. 21.STOCKHAM, J. T. G. 1972. Image processing in the context of a visual model. In Proceedings of the IEEE, vol. 60, 828–842.
  22. 22.TANAKA, T., AND OHNISHI, N. 1997. Painting-like image emphasis based on human vision systems. Computer Graphics Forum 16, 3, 253–260.
  23. 23.TUMBLIN, J., AND RUSHMEIER, H. E. 1993. Tone reproduction for realistic images. IEEE Computer Graphics and Applications 13, 6 (Nov.), 42–48.
  24. 24.TUMBLIN, J., AND TURK, G. 1999. LCIS: A boundary hierarchy for detail-preserving contrast reduction. In Proc. ACM SIGGRAPH 99, A. Rockwood, Ed., 83–90.
  25. 25.TUMBLIN, J., HODGINS, J. K., AND GUENTER, B. K. 1999. Two methods for display of high contrast images. ACM Transactions on Graphics 18, 1 (Jan.), 56–94.
  26. 26.WARD LARSON, G., RUSHMEIER, H., AND PIATKO, C. 1997. A visibility matching tone reproduction operator for high dynamic range scenes. IEEE Transactions on Visualization and Computer Graphics 3, 4, 291–306.
  27. 27.WARD, G. J. 1994. A contrast-based scalefactor for luminance display. In Graphics Gems IV, P. S. Heckbert, Ed. Academic Press Professional, 415–421.

Citation

MLA
Fattal, R., et al. “Gradient Domain High Dynamic Range Compression”. Proceedings of the 29th Annual Conference on Computer Graphics and Interactive Techniques, 2002, pp. 249–56, https://doi.org/10.1145/566570.566573.
APA
Fattal, R., Lischinski, D., & Werman, M. (2002). Gradient domain high dynamic range compression. Proceedings of the 29th Annual Conference on Computer Graphics and Interactive Techniques, 249–256. https://doi.org/10.1145/566570.566573
Chicago
Fattal, R., D. Lischinski, and M. Werman. 2002. “Gradient Domain High Dynamic Range Compression”. Proceedings of the 29th Annual Conference on Computer Graphics and Interactive Techniques, 249–56. https://doi.org/10.1145/566570.566573.
Harvard
Fattal, R., Lischinski, D. and Werman, M. (2002) “Gradient domain high dynamic range compression”, Proceedings of the 29th annual conference on Computer graphics and interactive techniques. ACM, pp. 249–256. Available at: https://doi.org/10.1145/566570.566573.
Vancouver
1. Fattal R, Lischinski D, Werman M (2002) Gradient domain high dynamic range compression. In: Proceedings of the 29th annual conference on Computer graphics and interactive techniques. ACM, pp 249–256

BibTeX

@inproceedings{Fattal_2002, series={SIGGRAPH02}, title={Gradient domain high dynamic range compression}, url={http://dx.doi.org/10.1145/566570.566573}, DOI={10.1145/566570.566573}, booktitle={Proceedings of the 29th annual conference on Computer graphics and interactive techniques}, publisher={ACM}, author={Fattal, Raanan and Lischinski, Dani and Werman, Michael}, year={2002}, month=July, pages={249–256}, collection={SIGGRAPH02} }
Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF