Tone Mapping

Tone mapping Tone Reproduction or dynamic compression are synonymous terms that denote the compression of the dynamic range of high contrast images ( high dynamic range images), so digital images with high brightness range. When tone mapping of the contrast range of a high- contrast image is reduced in order to display it on standard output devices.

Physiological background

In nature, a dynamic range (ratio of maximum and minimum luminance ) of about 109:1 occurs when one compares the sunlight with the starlight. The at any given time typically observed dynamic range is of the order of 1:10,000. The human visual perception solves the tone mapping problem as it is able to adapt to the ambient brightness conditions. On different absolute brightness conditions ( photopic, mesopisch, scotopic ) the eye does not respond linearly.

Many tone-mapping methods are based on knowledge about the human visual perception, since its goal is to compute a naturally as possible looking image. Play the most important role in the photoreceptors, the adaptation can be described by the Naka -Rushton equation as follows:

Here is the PR stimulus intensity, the maximum stimulus intensity, the light intensity and the light intensity that elicits half the stimulus intensity at the prevailing background intensity. A plurality of tone-mapping method based on an equation that is similar to this.

Method

There are many tone mapping operators, however, can be divided only into a few fundamentally different classes. So-called global operators use a function that assigns a value to each dynamic compressed HDR- value and is applied to each pixel. In contrast, this function will vary for each pixel based on local adaptation level of local operators. Frequency -based operators use a fundamentally different technique, in which the dynamic range of image regions is reduced depending on the spatial frequency. Finally, there are gradient-based operators that attenuate the brightness gradient of the output image for each pixel, to generate the LDR image ( image with low brightness range ).

Many operators expect that the values ​​of the output image as the luminance in a certain unit ( cd / m ) are calibrated. This is because the non-linear perception is taken into account of the absolute brightnesses; a daylight scene is therefore differently than a night scene. However, it is often possible to reconstruct the original lighting conditions directly from the HDR image by the histogram is evaluated. Most tone-mapping methods ignore the color perception largely and apply the new brightness value to all channels on.

Global operators

Global operators process the pixels of the output image independently. They are faster than other methods and can often be performed in real time. However, they are less well suited for scenes with a wide dynamic range, as they are more inclined to lose in very bright and very dark areas.

Many global operators are based on adaptation models where the background intensity must be known. This intensity can be estimated by the arithmetic mean of the pixel values ​​is calculated, however, the geometric mean is the preferred method.

The simplest global operator calculates the values ​​of the original image linearly down to the dynamic range of the LDR image. However, this method is insufficient because detail and contrast is lost.

Local and frequency-based operators

Local operators can process a large class of HDR images, as they can represent a wider dynamic range, without losing detail. They assume that human brightness perception does not adapt to the entire image, but only smaller regions.

To calculate the local brightness value for each pixel, a radial filter can be used which is applied to the neighboring pixels. However, this method leads to halo artifacts and contrast reversals near edges because there prevail to large differences in brightness within the filter radius. To work around this problem, different methods can be used:

• One possibility is, to the filter radius variieren.Der radius of the filter is doubled starting from the value of 1 until the pixels of the edge falsify the result, thus if the new average value from the old to a certain value from the original.
• Another possibility is to bilateral filtering. This is filtered by means of a radial filter, not only a function of the distance to the center pixel, but also a function of the absolute difference of the luminance values ​​. Pixels whose values ​​differ greatly from that of the central pixel, thus have little influence on the result. Durand and Dorsey used for both factors are Gaussian functions; Pattanaik and Yee use for the radial factor of a cylinder function and the brightness factor is an exponential function.
• Bilateral filtering tends to soften abrupt changes of luminance gradient. On the other hand, curvy areas and regions are not blurred sufficiently high gradient. Choudhury and Tumblin presented with the trilateral filtering an extension that also takes into account the brightness gradient.

For determining the optimum filter radius, a number of low-pass - filtered versions of the original image may be used.

Frequency -based operators divide the original image into a filtered HDR image with low spatial frequencies and an unfiltered LDR image with high frequencies, which are then combined. However, the filtered image can also be interpreted in such a way that each pixel provides a local adaptation value. Therefore, it is not always clear disconnect between local and frequency- based operators.

Gradient based operators

This class of tone mapping operators calculates the gradient of the output image and weakens them off.

Comparison

Tone- mapping operators differ in speed, strength and presence of artifacts, while maintaining image detail and the ability to compress HDR images with a very large dynamic range. Some studies deal with the comparison of tone mapping methods. The International Commission on Illumination, the Working Committee TC8 -08 made ​​to develop methods for the validation of tone mapping operators. When visual comparison of different operators, the difficulty arises that changes in parameters can have a large impact on the result.