How does the global tone mapping approach work?

The global approach computes a single mapping function for the entire image. It starts by calculating a log average of luminance values, establishes a middle gray, scales the input relative to this gray, and then compresses high luminances while enhancing low ones.

What is the difference between global and local tone mapping?

The global approach uses one function for the entire image, while the local approach adapts the luminance compression dynamically based on the brightness and uniformity of the surrounding pixels. This allows for better contrast enhancement in specific regions.

How does the local tone mapper determine the scale for adaptation?

The local tone mapper determines scale by finding the extent of an area around a pixel where brightness doesn't change significantly. This helps simulate how the human eye adapts to different brightness levels in different parts of the scene.

What are some other tone mapping approaches discussed?

Besides Reinhard's, the video mentions the bilateral filter, which is an edge-preserving smoothing technique, and gradient processing, which compresses gradients to preserve low-gradient details.

Is tone mapping reversible?

Reversing tone mapping is difficult, especially with local approaches, due to inherent information loss like quantization to 8-bit RGB. While global approaches might be more recoverable, local adaptations introduce ambiguity that requires inferential methods like optimization to approximate.

Is Reinhard's tone mapping considered the best?

Perceptually, gradient processing and Reinhard's approach are considered to work best among the three discussed methods. However, Reinhard's method is noted for being perceptually superior and is widely used in photography.

Key Moments

TU Wien Rendering #22 - Reinhard's Tone Mapper

Two Minute Papers

Science & Technology4 min read26 min video

Apr 30, 2015|5,277 views|59|4

Rendering Global Illumination ray tracing Vienna University Of Technology (College/University)tone mapping reinhard tone graphics tone mapping reinhard

Save to Pod

Key Moments

TL;DR

Tone mapping algorithms convert radiance to displayable RGB, with global and local Reinhard methods.

Key Insights

Tone mapping is essential for displaying high dynamic range (HDR) radiance values on standard displays.

The Reinhard tone mapping algorithm has both global and local variants, widely used in rendering and photography.

Global tone mapping applies a single mapping function to the entire image, based on a log-average luminance.

Local tone mapping adapts the mapping function based on neighborhood pixel information, preserving local contrast.

The local Reinhard method analyzes brightness uniformity around a pixel to determine the appropriate scale for compression.

Other tone mapping approaches include bilateral filtering and gradient processing, each with different strengths.

THE NECESSITY OF TONE MAPPING

The core challenge in computer graphics is simulating light transport and measuring radiance. However, raw radiance values, which can span an enormous dynamic range, cannot be directly displayed on standard output devices like monitors. These devices typically operate within a much smaller, predefined RGB color space. Therefore, a process is needed to convert these wide-ranging radiance values into a displayable format. This conversion process is known as tone mapping, and the Reinhard algorithm is a prominent and widely adopted solution for this problem.

GLOBAL REINHARD TONE MAPPING

The global version of the Reinhard algorithm applies a single, uniform mapping function across the entire image. It begins by calculating an average luminance for the whole scene. To avoid disproportionate influence from extremely bright areas, this average is computed in log space, providing a more representative baseline. This log average is then mapped to a defined middle gray value, establishing a reference point for brightness. Input luminances are then scaled relative to this middle gray, and finally, high luminances are compressed to fit within the displayable range, while lower luminances are enhanced.

LOCAL REINHARD TONE MAPPING

In contrast to the global approach, local tone mapping adapts the mapping function for each pixel based on its surrounding neighborhood. After computing the overall log-average luminance, the algorithm determines a local average brightness and an appropriate scale for each pixel. This scale represents the area around the pixel where brightness remains relatively uniform. By considering this local uniformity, the algorithm can better simulate how the human eye adapts to different brightness levels, enhancing contrast in areas where details might otherwise be lost due to compression.

DETERMINING THE LOCAL SCALE

A crucial aspect of the local Reinhard method is determining the 'scale,' which signifies the extent of an area around a pixel where brightness does not change significantly. This is achieved by analyzing concentric circular regions. If small and large circles around a pixel show similar smoothness in brightness distribution, the scale might be too small, suggesting larger circles can be used. Conversely, if a larger circle encompasses significant brightness variations (like crossing a window or a silhouette), it indicates an effective boundary. This process helps identify regions where contrast enhancement is most beneficial, mimicking human visual adaptation to uniform and non-uniform areas.

ADVANTAGES OF LOCAL ADAPTATION

The local tone mapping approach offers significant advantages in preserving image detail and contrast, especially in scenes with strong variations in illumination. By analyzing the local brightness distribution and adapting the tone mapping curve accordingly, it can differentiate between uniform regions and areas with sharp discontinuities. For instance, in a scene with detailed angel figures on a wall next to a uniform window, the local method can enhance the contrast of the figures while appropriately mapping the window's brightness. This localized adaptation more closely mimics the capabilities of the human visual system.

ALTERNATIVE TONE MAPPING STRATEGIES

Beyond the Reinhard algorithm, other tone mapping techniques exist. The bilateral filter, used in computer vision, offers edge-preserving smoothing, conceptually similar to how Reinhard's scale estimation avoids smoothing over bright edges. Fatal gradient processing views images as height functions and compresses gradients, preserving details in dimly lit regions by reducing the range of steep slopes. While these methods share the goal of tone mapping, they employ distinct strategies. The Reinhard method is often preferred for its connection to photographic techniques, while gradient processing allows for speed, though perceptual quality might vary.

THE EVOLUTION AND PERCEPTION OF TONE MAPPING

The Siggraph 2002 conference was a significant year for tone mapping, with multiple important algorithms presented. While early methods focused on speed, modern hardware capabilities reduce the urgency for extreme approximations. Perceptual quality is now a primary driver, making algorithms that closely match human vision more desirable. Reinhard himself continued to research and publish extensively in this field, cementing his status as an expert. Researchers can trace the development of tone mapping by examining cited papers, offering insights into various approaches and their effectiveness over time.

REVERSIBILITY AND INFORMATION LOSS

Reversing the tone mapping process to recover original radiance values is generally not perfectly possible, primarily due to inherent information loss. Quantization, when mapping 24-bit HDR values to 8-bit RGB, irretrievably loses precision. Local tone mapping approaches further complicate reversibility because the applied compression often depends on context-specific neighborhood information. Without knowing this local adaptation history, it becomes difficult, if not impossible, to accurately reconstruct the original high dynamic range. Global tone mapping, being a uniform process, offers a better, though still not perfect, chance of some degree of reconstruction.

Mentioned in This Episode

●Organizations

●Books

●Studies Cited

●Concepts

Reinhard's Tone Mapper: Global vs. Local

Practical takeaways from this episode

Do This

Understand that global tone mapping applies a single mapping function to the entire image's luminance values.

Compute a log average to establish a baseline brightness for the scene.

Map luminance values to a middle gray value to set the range.

Compress high luminances and enhance low luminances using a non-linear mapping function.

For local tone mapping, adapt the function based on neighboring pixel brightness to account for local average brightness.

Consider the scale of uniform brightness regions around a pixel to determine local adaptation.

Recognize that local tone mapping can enhance contrast in uniform regions like angel figures on walls.

Avoid This

Do not simply average luminance values without using a log average, as this gives unproportional weight to large values.

Avoid using a uniform compression function across the entire image if local adaptations are needed for contrast and detail.

Be aware that making the scale discs for local adaptation too large can lead to erroneous results.

Do not expect full reversibility of the tone mapping process, especially with local approaches, as information can be lost.

Common Questions

Reinhard's tone mapper is an algorithm developed in 2002 that adjusts the luminance values of an image to make high dynamic range content visible on standard displays. It exists in both global and local variants.