|CCD Instruction Manual||Al Kelly Home Page||Latest Images||Galaxies||Nebulae||Star Clusters||Solar System||Links|
Color CCD Imaging with RGB and CMY Filters
Two new techniques in CCD imaging make true color image production easier and better. These developments are using solar-type stars as color-calibration standards for red, green, blue (RGB) imaging, and using calibrated high-transmittance cyan, magenta, and yellow (CMY) filters for color imaging. Both techniques work well with another recently developed technique: that of combining a fourth high-quality white-light image with the three tricolor images. Taken together, these three techniques promise a virtual revolution in CCD color imaging.
The True-Color Problem
Color is problematic in astronomy because the color-sensitive cone cells in the human retina do not respond to faint light. As a result, astronomers are not familiar with the color appearance of deep-sky objects in the same way we are with the colors of trees, animals, and people. To see color, we resort to long exposure color photographs or recombine CCD images shot through filters. Standard "tricolor" imaging uses red, green, and blue filters -- and the problem has been how to combine them to produce an accurate color rendition. Our "imaging team" has found that Sun-like stars (called "solar analog" stars) allow us to calibrate our red, green, and blue images to produce accurate color images. In addition, we have found that you can calibrate cyan, magenta, and yellow filters using Sun-like stars, and also create accurate color images.
Conventional RGB filters sample the light from astronomical objects in three wavelength ranges that correspond to the retinal cells in the human eye. These are red (600 to 700 nm), green (500 to 600 nm) and blue (400 to 500 nm). CMY imaging records these wavelength ranges in pairs. The cyan filter transmits blue and green, the magenta filter transmits blue and red, and the yellow filter transmits green and red.
CMY filters transmit more light and provide better coverage of the spectrum than RGB filters do. The photon throughput of CMY filters is approximately twice that of RGB filters, so exposure times for luminance (brightness) purposes can be halved. It should be recognized, however, that the signal-to-noise ratio of chrominance (color) data taken by RGB filters is about 18% better than that taken by CMY filters. Another consideration is that if the transmission curves of a set of RGB filters do not overlap, important spectral lines may be lost between two color bands. For example, the 500.7 nm line of doubly ionized oxygen resides halfway between the peak values of blue and green coverage and is poorly sampled by some RGB filter sets. Since this line is a significant part of the flux from many colorful objects, such as planetary nebulae, the full sampling of this line by the cyan filter in a CMY set is very important for true color representation.
[NOTE: the highlighted section above was revised on 9/8/02 by one of the authors (Al Kelly)]
A Team Effort
Our "imaging team" (consisting of Richard Berry, Chuck Shaw, Ed Grafton, and Al Kelly) developed the new techniques in a roundabout way. Like everyone who does CCD imaging, we had taken literally thousands of unfiltered images, but then we explored tricolor imaging. Richard had written a program (Color245) for producing color from tricolor and Al, Chuck, and Ed were shooting tricolor with their telescopes, but RGB imaging with CCDs is made difficult by the low blue sensitivity of front-surface CCDs.
We were looking for a solution when we saw the great results that Robert Dalby in the UK and Kunihiko Okano in Japan had obtain by combining standard RGB tricolor images with a fourth image (the luminance, or white-light image). Dalby and Okano used PhotoShop to create a red, green, blue color composite, and then replaced the original luminance layer with a deep image taken without a filter.
Al and Ed shot some white-light images and started experimenting with luminosity replacement in PhotoShop, and Richard rewrote his Color245 software. To enhance the color output, Color245 already converted the RGB images into hue, saturation, and brightness, altered the image brightness and color saturation, and then converted the pixels back to RGB. Richard added one more wrinkle: after he converted a set of RGB imaging into hue, saturation, and brightness, he replaced the computed brightness with the brightness of the white light image. Al and Ed were soon shooting their best color ever using this powerful "quadcolor" technique.
We were troubled, however, by difficulties in knowing whether our color images were "accurate" in their color. We agreed on a definition of true color rendition: "approximately what the human eye would see if it were sensitive enough to see all the color in the light from celestial objects," and we also agreed that we need ways to eliminate the ugly green and brown sky backgrounds that light pollution produces.
Although "what we would see" looks like a clear definition, the truth is elusive. Human color perception is a complex, nonlinear product of the eye-brain system. Even people with normal vision see colors somewhat differently, the eye loses color perception in low-light conditions, and no person has ever been privileged to see deep-sky objects up close, with fully active color vision. Since red, green and blue-sensitive cones in the human eye have different thresholds of sensitivity, we realized (and have noticed visually) that the eye sees different color balances under different low-light conditions. Our definition is adapted to what nature provides, even though it may be in small amounts.
A problem that especially bothered us was that we suspected that the passbands of our red, green, and blue filters did not overlap, so that important lines in the spectra of planetary nebulae like M27 went undetected. We decided to try cyan, magenta, and yellow filters because each filter passes a broad sweep of spectrum, so that no spectral lines would be missed.
With a suitable standard in mind, we saw four issues:
* Conversion of CMY data to RGB data. A theoretically simple relationship exists between RGB filters and CMY filters. CMY filters are called subtractive filters because they each block an RGB portion of the spectrum. Cyan blocks red, magenta blocks green, and yellow blocks blue. However, this simplicity partially fails in practice, because the pass bands vary slightly from filter set to filter set and do not map exactly in accordance with color theory.
* Spectral sensitivity of CCDs. The CCDs that amateur astronomers use have greater sensitivity to red light than to green or blue light. Specifically, for the Cookbook 245 and SBIG ST6 cameras that we use, the ratio is roughly 4:3:2. To "balance" the colors, the red, green, and blue images must be "weighted" differently when they are composited into a color image.
* Correction for Atmospheric Extinction. The Earth's atmosphere transmits RGB wavelengths unevenly. Wavelengths from 400 nm to 700 nm are transmitted with approximately equal efficiency at the zenith, but as the angle of elevation of an object approaches the horizon, blue transmission is extinguished much more than red. Again, green extinction is somewhere in between. True color rendition requires that this factor be taken into account.
* Correction for Sky Background Color. Depending on where you observe, the light of the night sky distorts the colors of objects seen through it. Particularly in light polluted areas, this is a major factor in true color rendition. The color of the foreground sky near cities is frequently a dirty greenish-blue to yellow. Even on dark observatory mountain tops, the foreground sky adds a reddish tinge to the true colors of objects.
To derive solutions which could be implemented by any careful CCD imager, we had to develop a method of measuring the pass-band characteristics of RGB and CMY filter set and determining the RGB spectral response of a CCD camera. After a series of rather confusing efforts shooting daytime test scenes, which included photographic gray cards and colorful snips of paper and piles of fruit, we managed to make reasonably accurate color images of our test setups using both RGB and CMY filter sets.
For color calibrating astronomical images, Richard suggested that sun-like spectral type G2 class V stars ought to be the target, since they provide pure white targets in the sky. We caught fire with this idea and received valuable assistance from Brian Skiff, who was kind enough to send us a long list of "solar analog" stars spread evenly around the sky. We reduced this list to 21 stars easily found on many of the printed and electronic sky maps available to amateurs. (See list below)
To capture the necessary data, we conducted numerous imaging sessions of stars from this list, capturing images through each filter from about 15 degrees elevation above the horizon to near the zenith. Generally, we captured a full set of filtered images every 5 to 10 degrees of elevation as the star rose or set. There were seven calibrated and averaged images to each set: one with the infrared blocking filter in place to get white-light star data, and one through each of the R, G, B, C, M, and Y filters. The infrared blocking filter was always in place, as it was for all our filtered deep-sky images. We were also careful to assure that all images were the same exposure duration and that they were short enough to avoid saturating any pixels, which would also skew the data. Exposure times were usually very short, from .075 seconds to 0.999 seconds, depending on the brightness of the object star.
While Chuck, Ed and Al were capturing data, Richard added a star photometry function into his QColor software so that we could measure the CCD's response to the filtered starlight. We also needed the right equations for converting CMY data to RGB data, equations which would take into account the vagaries of different CMY filter sets and correlate them to the way RGB filter sets see the universe. Chuck and Richard came up with a matrix transform to convert CMY image data to RGB, and the last roadblock was removed. Within two days, Richard produced a new version of QColor that applied the conversion equations and allowed us to produce the first true color WCMY images.
After a complete set of star data was taken, we did the color calibration by measuring the brightness of the star in each filter at each elevation. These data were converted to magnitudes. When these are plotted against the atmospheric path length (called the air mass), the data for each filter should result in a straight line. We used a least-squares solution to obtain the brightness the star would have at the zenith and the atmospheric extinction coefficient, measured in magnitudes per air mass in red, green, and blue light.
The measured brightness of a Solar-analog star through the different filters told how our CCDs responded to pure white light as passed by each filter. Taking the CCD's response to white starlight through the infrared blocking filter as 100%, we saw, due to uneven spectral sensitivity, approximately 40% of that response in red, 30% in green, and 20% in blue (the CCD's response is the product of the filter transmission and the CCD's quantum efficiency, so they do not sum to 100%). The cyan, magenta, and yellow responses were about 40%, 50%, and 70%, respectively. These figures are close approximations of the average data from our systems.
We next calculated weight factors for the filters. The red weight was Rwt = 100/40 = 2.50; the green weight was Gwt = 100/30 = 3.33; and the blue weight was Bwt = 100/20 = 5.00. This means that if we shoot a deep-sky object with equal exposures in red, green, and blue, we can simply multiply the red image by 2.50, the green image by 3.33, and the blue image by 5.00 to balance their pixel values so that a white star produces a white image on the computer screen.
Since the cyan, magenta, and yellow filters each pass two of the three primary passbands( red, green, and blue), we can calculate the required weights for the RGB data generated by the cyan, magenta, and yellow filters. This assumes that the passbands of the CMY filter sets are close to ideal passbands, but our tests have shown that the assumption appears to be justified with most filter sets. To obtain weights for the RGB data and normalize it to red: red weight = red data/red data = (M+Y-C)/(M+Y-C) = 80/80 = 1; green weight = red data/green data = (M+Y-C)/(C+Y-M) = 80/60 = 1.33; blue weight = red data/blue data = (M+Y-C)/(C+M-Y) = 80/20 = 4.0.
Finally, to convert the cyan, magenta, and yellow images into red, green, and blue images, we combine them, pixel by pixel, as follows:
R = (Y + M - C) * red weight
G = (Y + C - M) * green weight
[NOTE: the highlighted section above was revised on 5/21/00 by one of the authors (Al Kelly)]
To check our empirical results, Richard developed computer software and finds that for well-matched filters this method gives excellent color images. Not surprisingly, the more the RGB and CMY filter sets depart from ideal, the less accurate the color. All told, we developed six different methods for calibrating and converting CMY image sets into RGB image sets. We found that the relatively simple method described above produces the best results for the least amount of work.
To further check our methods, Dr. Dick Morris of the Johnson Space Center Earth Science and Solar System Exploration Division generously made high-resolution spectrophotometer scans of our filters. Richard has used this in conjunction with high-resolution stellar spectra (thanks to Dr. James Kaler and Dr. George Jacoby), the published quantum efficiency of our Texas Instruments TC245 CCDs, and standard values for the transmission of the atmosphere. His models agree with the image test data, giving us confidence that our simple method allows any amateur to acquire and color-calibrate RGB or CMY images to produce accurate CCD color images.
After conversion into an RGB image set, two steps remain: correcting for atmospheric extinction and removing the foreground sky color. Richard's models show that when we obtain a good data set, our empirical results match the standard atmospheric extinctions used at major observatories. For consistency, we recommend multiplying the red, green, and blue images by the "typical" extinction given in the table below, to restore the light lost to the filtered images by extinction.
Correction for atmospheric color is the final step in the process of producing true color. Since the foreground sky is light added to the image color, we simply measure the foreground sky in each image and subtract a number chosen to make it the same low pixel value over the whole image. This is justified because the true cosmic background light is only about 1/200 as bright as a rural night sky, effectively a very dark shade of gray. QColor has a "color adjust" function that allows correcting for atmospheric extinction and the foreground sky color in one simple operation.
We emphasize that making accurate color images, whether using RGB or CMY, is somewhat involved, but can be accomplished by any careful imager. The resulting images are all the more satisfying because the color you see on the computer screen is not the arbitrary result of pushing the color slide bars in PhotoShop, but the result of precise calibration against a known color standard. However, there are a number of factors that anyone contemplating color imaging must consider:
* Our methods show how deep-sky objects would look to the eye if they were a thousand to a million times brighter. Under low-light conditions, the eye is extremely nonlinear, so the subtle colors that you see with your telescope may not match the color these methods reveal.
* The high red sensitivity of some CCDs produces "white light" images that are too heavily influenced by red light. If these images are used to supply the luminance in an RGB or CMY image, the red parts of the image will appear too bright relative to blue parts of the image. Usually this is not a significant problem, but you should be aware of it.
* Your filter set must include a good infrared-blocking filter used with other filters when taking images. Anything more than a minor leakage of infrared light will cause "bleaching" of colors and throw off the color balance. If you can, use a suitable dichroic infrared blocker rather than a dyed-glass filter.
Astronomical pictures are seldom accurate in color. Instead, they are subject to technical problems like reciprocity failure and color trapping on high-speed printing presses, as well as the whims of amateur photographers and magazine art directors. If you take care in acquiring, calibrating, and producing RGB or CMY color images, your results will be more color accurate than the pictures you see in magazines, books, or on the web.
Our "imaging team" has tested our new methods of shooting and color-calibration with a variety of telescopes and a variety of sites. Even though Al uses a CB245 camera and Ed uses an SBIG ST6 camera, and Al's and Ed's CMY and RGB filter sets are different, the color renditions have been remarkably similar. Most of Ed's images have been made from light-polluted skies within the city limits of Houston, Texas, while Al's imaging locations have varied from slightly less light-polluted suburban skies to the profound darkness at the Texas Star Party in the Davis Mountains of West Texas. We feel that this commonality of results under widely varying conditions provides very strong support for our testing, image conversion, and image correction method.
The images below of the Sombrero Galaxy made by Al at the 1998 Texas Star Party demonstrate the effective conversion of CMY data to RGB data. Both the WRGB and WCMY images were processed using the conversion and filter weights derived by earlier star test data. Atmospheric extinction and color were corrected equally for both so that a true comparison of color balance could be made. The nearly identical images show that WCMY passes muster, especially since the total exposure time of the images for the WCMY composite was 9.7 minutes, while the WRGB composite required a total of 21.3 minutes!
We believe that careful color calibration of a CCD camera using standard "solar analog" stars, and accounting for variations in filter sensitivity, atmospheric transmission, and sky background color make it possible for any amateur astronomer to produce color images of exceptional color fidelity. Both the classic RGB and the new CMY filter sets are capable of yielding true-color deep-sky CCD images even at light-polluted observing sites. It is the best answer short of turning out the lights!
In conjunction with this project, Richard has produced data conversion, modeling, and image processing software to support the full range of needs for filter calibration and creation of WCMY composites. At present, this software supports Cookbook camera image formats. Interested parties should visit Richard's website for further information.
We also want to recognize important contributions to color CCD imaging by Kunihiko Okano, of Japan, and Robert Dalby, of England. Using Adobe PhotoShop, Okano and Dalby independently developed methods of layering RGB tricolor images with high-quality luminance images to produce color images with greater detail and impact. Their work was an inspiration to us.
Our thanks to Brian Skiff for providing a list of main-sequence G2 stars for use as calibration standards. Brian notes that planetary astronomers uses these stars as "substitute suns" to calibrate photometric observations of comets and asteroids.
Readers are invited to visit the authors' websites to see a growing collection of WCMY images and to get further information.
Table 1: Solar Analog Stars
Table 2: Extinction Correction Factors
EL.ZD..AM.. Rxc... Gxc... Bxc
90 00 1.000 1.000 1.000 1.000
80 10 1.015 1.001 1.002 1.003
70 20 1.064 1.005 1.010 1.014
60 30 1.155 1.013 1.025 1.035
55 35 1.221 1.018 1.036 1.050
50 40 1.305 1.025 1.050 1.070
45 45 1.414 1.034 1.068 1.097
40 50 1.555 1.046 1.092 1.132
35 55 1.743 1.063 1.125 1.180
30 60 2.000 1.085 1.172 1.249
25 65 2.365 1.118 1.242 1.356
20 70 2.923 1.170 1.356 1.535
15 75 3.862 1.263 1.574 1.892
In this table, EL is the elevation angle of the object above the horizon, ZD is the distance of the object from the zenith, AM is the air mass, or atmospheric thickness relative to the zenith, and Rxc, Gxc, and Bxc are extinction correction coefficients for red, green, and blue filters.