English
日本語
简体中文
There is a perception that color precision can only be achieved with 3-chip prism-based (3-CCD, multiple-chip prism, etc.) video cameras. Because of advances in sensor design, knowledgeable camera designers can now get excellent color images from Bayer patterned image sensors. These images are so good that all consumer digital SLR cameras use a single chip sensor configuration.
Background
There are three common ways of spatial color detection with cameras:
1. Use of a mosaic color filter array (CFA) on a single image sensor. Different primary filters are distributed over the pixels of the sensor. Each filter has a specific (band) transmission spectrum. There are several primary filter types (Y,C,M or R,G,B, etc) and arrangements (stripe, Bayer, pseudo random).
The Bayer RGB CFA (Figure 1) is commonly used in digital still cameras, camcorders, and scanners. The Bayer filter pattern is 50% green, 25% red and 25% blue. The reason there is more green is that both silicon and the human eye are most responsive around the green wavelengths.
Figure 1. Bayer Mosaic Color Filter Array Configuration
2. Use of a beam splitter to project the same image on 3 image sensors. Each sensor has its own primary color filter. Then data from all 3 of the image sensors are combined to make a color image. 3-CCD prism-based cameras are a widely known implementation of this technique shown in Figure 2.
Figure 2. 3-CCD Prism-Based Color Configuration
3. Use of the wavelength dependency of photon absorption in silicon and separately collect the charge at different depths in the sensor. This way, each pixel directly has its own ‘RGB’ value (Figure 3). On the sensor level, it is extremely challenging to ensure the separation of the colors and prevent leaking electrons from one depth to the other. There is R&D work on this technique, but it has not been perfected for production use.
Figure 3. Color Imaging Using Photon Absorption in Silicon
For this discussion, we will consider the advantages of color cameras utilizing single chip digital image sensors with a Bayer CFA versus 3-CCD prism-based industrial cameras in terms of image quality, costs, and robustness (see the next blog for costs and robustness considerations).
Image Quality
Some of the frequently mentioned criticisms of single chip color cameras compared to 3-chip cameras include reduced resolution, reduced sensitivity, and lesser color fidelity. Let’s look at these in more details:
Resolution
The mosaic filter approach lowers the spatial resolution available by roughly 30% compared to a monochrome sensor or 3-CCD approach in which all of the photons are used for the image.
The resolution problem is very easy to overcome as getting more pixels is not a problem these days. Also, very smart processing is available to get the most out of the sensor to maintain sharpness, etc. This is why single chip solutions are no problem for DSLR cameras.
Sensitivity
With a single chip color cameras, not all of the pixels are sensitive to all colors so not all of the light contributes to sensitivity as it does with 3-chip cameras.
Even though part of the light is lost, the noise level in the latest sensors has gone down so much that they are still good enough for many applications. Even in low light levels, the SNR is more than acceptable.
Also, beam-splitting prisms often include absorption filters to provide more precise color separation. This means the sensitivity advantage is not as great as it would seem.
Color Fidelity
With 3-CCD technology, a specific R, G, and B value is captured for each pixel and is therefore expected to produce higher color precision. In practice, the color fidelity relies on a properly aligned and precise prism to prevent color fringing and chromatic aberration.
With a Bayer pattern image sensor, each pixel collects data for one of the three Bayer primary colors (RB, GB, or BB). Alternatively put, the image consists of three subsampled color images (RB, GB, BB). For each of the three, the gaps are filled by interpolation – this is often called demosaicing. With accurate demosaicing artifacts in color images can be prevented, particularly near sharp edges
The latest filters provide excellent bandpass transmission separating the colors with a high degree of precision providing stable performance and minimal crosstalk. Extremely accurate color reproduction can be achieved with color cameras utilizing 1 image sensor.
Read why 3-CCD solutions are not good enough for global security applications in our next blog post.
Unmentioned in the text is the fundamental flaw in 3-chip design that limits its color fidelity. If “all of the light contributes to sensitivity” then dichroic beam splitters must be used. They in turn entail that there is little or no overlap in the R,G,B spectral sensitivities. Author Gretchen Alper might think that is good, for she speaks of “excellent bandpass transmission separating the colors” for some Bayer pattern sensors, but it is actually color-scientifically bad. The human cone responses have great spectral overlap, and any R,G,B sensors that are linear combinations of the cone responses, and positive valued, will also have great spectral overlap. Deviations from that “Luther Condition” imply deviations from human-agreeing color capture, that no post-capture manipulations can undo.
Thus successful Bayer pattern digital cameras (for human picturing) have their R,G,B spectral sensitivities strongly overlapping (if not exactly satisfying the Luther Condition). Efficient 3-chip designs can’t have that spectral overlap, which is the real reason they have limited color fidelity.
Dear Dennis,
Thank you for your reaction on our post. For this reason we have decided to focus on Bayer-pattern color sensors and cameras.
Regards, Gretchen Alper