Bayer filter

A Bayer sensor is called a photo sensor - is coated with a color filter, which usually consists of 50% green and 25% each of red and blue - similar to a checkerboard. Green is in the area of ​​assignment ( thus the resolution capability) privileged because green (or the amount of green in shades of gray ) in the human eye makes the greatest contribution to brightness perception, and thus also to the contrast and sharpness of perception: 72% of the brightness and contrast perception of shades of gray is caused by the amount of green, red, however, makes only 21% and blue only 7%. In addition, Green, as the middle color in the color spectrum, the one for which lenses usually deliver the highest image quality (sharpness, resolution).

According to this concept, the Bayer matrix (English Bayer pattern ) almost all common CCD sensors work in digital photo and video cameras. The concept of this type of sensor is in contrast to the concept of Foveon X3 direct image sensors. See also Super CCD sensor.

The " Bayer matrix " or " Bayer filter " is named after its inventor, Bryce E. Bayer, who on March 5, 1975, the patent on behalf of the Eastman Kodak Company in the United States filed (patent US3971065: Color imaging array. ).

Operation and design

The light-sensitive cells of a single photocell on the semiconductor can only detect brightness values ​​. To obtain color information, a color filter is applied minute in one of the three primary colors red, green and blue before each cell. The filters are applied, for example, in the odd rows in the sequence red - green and in the straight line in the sequence blue-green. Each color element (pixel ) corresponding to only provides information for a single color component at this point, so that a complete picture of the same dimensions, the respective adjacent pixels of the same color have to be used to color interpolation. Thus, 50 % of the pixels are calculated for green, blue and red are at 75% of the surface ( or in a line 50%, and 100% in the next line of the line) that must be filled by calculation. In the color interpolation, it is assumed that there are only slight differences in color in the image between two adjacent pixels of the same color, and hence the gray scale values ​​of the pixels are not stochastically independently. This of course does not apply to any motif. Therefore, the Bayer sensor has under consideration an artifact-free image, strictly speaking, only a quarter of the apparent resolution.

In addition, such sensors have almost always more pixels that are located at the edge of the sensor surface and blackened, as a rule, in order to determine during the operation under exposure, the temperature-dependent noise of the sensor and to take account computationally, eg for calculating a compensation value ( "offset" ) for the evaluation of the other pixels. In addition, these pixels can also, for example, to detect extreme overexposure, use for example by too long integration time ( = exposure time) of the sensor elements. For normal camera user but they are of no significance, since the adjustment process runs automatically and is possibly realized depending on the model already directly on the sensor.

Interpolation

The above-mentioned interpolation (English demosaicing ) can be performed in various ways. Simple method to interpolate the color value of the pixels of the same color in the neighborhood. Since this procedure is particularly problematic at edges perpendicular to try other methods, especially preferred along rather than to perform the interpolation of vertical edges. Still other algorithms rely on the assumption that the color of an area in the image is relatively constant even under changing light conditions, so that the color channels are highly correlated with each other then. Therefore, first, the green channel is interpolated after the red and interpolate the blue channel, so that the respective ratios of colors red-green or blue-green are constant. There are other methods that make different assumptions about the image content and on the basis of this attempt to calculate the missing color values. 5 x 5 matrix, for example, filtering a smoothed image is generated, which is then honed again.

Problems ( artifacts, often referred to as interpolation artifacts ) can arise if the assumptions made by the algorithm are injured in an actual recording. For example, the above-mentioned, exploited by many higher algorithms assume that the color levels correlate not more consistently accurate when due to chromatic aberrations of commercially available lenses the color planes are displaced in the border regions against each other.

Another challenging problem are striped pattern with a stripe width corresponding to approximately that of a single pixel, for example, a picket fence in a matching distance. Since a signal generated by such a motive Bayer raw image pattern could have been generated by both horizontal and by vertical picket fences ( different colors), the algorithm has to make a decision whether it is a horizontal or a vertical structure, and which color combinations should be evaluated as such plausible. Since the algorithm does not have human motive - world knowledge, it is then often to random decisions, thus to wrong decisions. For example, such a picket fence then erroneously as a random mixture of horizontal and vertical portions, so like a maze, is shown upside down.

Consider, for example, a motif neckline, within which all the red pixels are lit and the green only those in the red columns. For this illustration, the following motives alike would fit:

  • A vertical white picket fence with slats in front of black background
  • A vertical picket fence with yellow slats in front of black background
  • A vertical picket fence with white slats against red background
  • A vertical picket fence with yellow battens against red background
  • A horizontal slat fence with red slats in front of green background
  • A horizontal slat fence with red slats against yellow background
  • A horizontal slat fence with purple slats in front of green background
  • A horizontal slat fence with slats violet against yellow background
  • As well as all conceivable picket fence or lattice motifs that are any mixture of the above options.

In this simple theoretical example, an algorithm could, for example the variant with the lowest overall color preference, thus vertical slats white against black backdrop accept. In practice, however, the structure alignments agree hardly exactly with the Bayer grid agree, that in such a picket fence motif no black and white option to choose would be, but several alternatives with similar Farbigkeits plausibility compete for the privilege, and it comes to random decisions.

These problems are, however, the more toned down than the resolution of modern sensors could reach the resolution of lenses or even exceed, especially when using zoom lenses, or lenses, the entry price category. Since the resolution limit of lenses (as opposed to sensors) is not fixed, but is defined as a contrast border, this means that finely - detailed design cutouts with a high artifact inclination of such lenses only with a very low contrast can be imaged on the sensor. Thus, interpolation artifacts such motif cutouts have a very low contrast and have a less disruptive effect.

Processing example

An example of a Bayer image reconstruction software. The images are shown enlarged by a factor of 10 because of the representation.

SW- sensor image overlaid with virtual color filters

Preprocessed image using software filter

Fully processed image reconstruction including

Alternative developments

Kodak had experimented with various pixel arrangements with additional "white" pixels. Even Sony had installed in some models "white" pixels and, for example at the 2003 Sony DSC -F 828 an image sensor with two different shades of green used ( RGEB = red (red) / green (green) / emerald ( emerald green) / blue (blue ) ).

A Bayer variant has been developed in which the two green pixels in a 2 × 2 blocks each have different color filters ( for slightly different greens ). This version was installed, for example in the Canon EOS 7D and other manufacturers at times.

A different approach is presented with his 2012 put on the market Fujifilm X -Pro1 Fujifilm: The RGB pixels were in another relationship and a different arrangement ( xtrans ) distributed on the sensor, so that on the one hand the track next to the pixel repetition extended on average, but also occurs each color pixel in each row and column. Since red and blue pixels are no longer exactly 2, but on average 2.23 units of length are now removed from their next neighbors of the same color, the resolution of the red and blue plane is about 10 percent is reduced, however, paradoxically but also the green resolution. For every green pixels within a green 2 × 2 square here has indeed continue to be as the Bayer pattern, always exactly 4 green neighbors, but now unevenly distributed: 2 more and 2 more distant.

Since the mathematical research on color pattern interpolation algorithms out regularly from a classic Bayer pattern, such alternative color pattern ideas from the quality of recent algorithmic approaches can not benefit and are therefore in the quality of implementation in a full color image by raw conversion software disadvantaged.

109769
de