CCD Image Processing 

    Creating striking color CCD images using LRGB composites has become the image processing technique of choice for those who image the night sky.  The theory and practice of the LRGB method have been particularly well presented in two publications:  The Handbook of Astronomical Image Processing by Richard Berry and James Burnell discusses the theory and The New CCD Astronomy by Ron Wodaski covers the practical aspects.  The websites of Robert Gendler and William McLaughlin contain excellent 'how-to' discussions of the LRGB layers method which I found helpful when I began to experiment with the  process.  

   The L (Luminance) image is obtained in unfiltered light using a high resolution CCD camera, such as the ST-10E binned at 1X1, to obtain a grayscale image.  Deep exposures are required to give a high signal-to-noise ratio (S/N) in the L image.  Red, Green and Blue images are obtained using RGB color filters with the camera binned at 2X2. Alternatively, one uses a different CCD camera with large pixels and/or higher sensitivity (such as the FLI-DM).  These measures are necessary to overcome the weaker signal of the filtered light which results in low S/N of the RGB images.  The four images, L;R;G &B are combined using image processing software.  The final LRGB composite reveals the detail of the Luminance image as well as the color information present in the noisier, lower resolution RGB images.  This is due to the fact that our visual processes perceive detail almost exclusively in the Brightness or Luminence component not in the Color components of the image.  Moreover, the eye ignores noise in the Hue and Saturation (Color) components.  Therefore, LRGB compositing produces a color image with a lot of detail...the Holy Grail of astronomical color imaging.   The same principal is used in the Component Video feature present in digital television images.

Processing the Luminance Images:

   High signal-to-noise images are obtained by combining many shorter exposures to yield long total integration times.  Typically, 12 ten min. exposures (or 24 five min. exposures) from the ST-10E, binned at 1X1, are combined to give a total integration time of 2 hours.  Ten (10s) flat fields are obtained at the conclusion of an imaging session.  Sets of 10 to15 raw dark frames at each exposure time are kept as a library which is turned over every few months.  On summer evenings, when external water cooling is required to maintain the camera at -20 degrees C, better dark frames are obtained when they are taken immediately after the imaging session.  The raw images are processed as follows:

  1. The raw FITS images are reduced, using master dark frames and dark-corrected master flat field frames, with AIP4WIN software.  (Master Darks & Flats are created using median combine.)
  2. The reduced images are aligned as a folder in CCDSoft.  The aligned images are then combined.  I use average-combine for most objects.  Dimmer objects are combined using the SUM.  Any remaining hot pixels in the final image are removed using the Repair Menu in CCDSoft.
  3. CCDSharp (SBIG) is then used to apply a deconvolution algorithm to sharpen the image and reveal hidden detail.  Anywhere from 5 to 9 iterations are used, depending on the S/N of the image.  The deconvolved image is saved as a FITS file.
  4. This image is converted to a 16-bit TIFF file and imported into Adobe Photoshop 6.0.  Black and white point adjustments are made to the Histogram to exclude pixels outside the useful brightness range.  Non-linear histogram stretching using Curves further shapes the histogram to emphasize or de-emphasize certain brightness levels.  This operation is also applied to isolated areas in the image using the Selection and Feathering tools.  Unsharp Masking is used to sharpen the highlights and Gaussian Blur to smooth the dimmest, and therefore noisiest, regions of the image.  At this point the image is converted to and saved as an 8-bit TIFF file.  Judicious use of the Curves, Unsharp Mask and Gaussian Blur tools is used additionally after this point, as well as the Burn and Dodge tools.  Remaining blemishes (including blooming) are removed with the Clone Stamp tool and Dust & Scratches tools.  Bright, bloated stars can be rendered more aesthetically pleasing by selecting the star (or stars, using a selection mask Layer) and applying a variety of tools including the Minimum FILTER, Unsharp Mask and Curves.  The final uncropped 8-bit image is saved as the L component for LRGB layering.      

Processing the RGB Images:

    RGB exposures are taken with the FLI-DM binned at 1X1(with R:G:B exposure ratios determined from white balancing). Three sets of 5 exposures for each of the three filters are collected.  Five flat fields (10s) are obtained with each filter at the conclusion of the imaging session.  Sets of 10 dark frames are kept as a library with a 3 month turnover.  The raw RGB images are processed as follows:

  1. The 5 raw FITS images in each set of R,G & B exposures are first dark-subtracted and flat field corrected in AIP4WIN.
  2. The reduced images in each set are aligned then combined as a folder in CCDSoft.  Median Combine is used to avoid color speckling. 
  3. The resulting three R,G, and B med-combined images are aligned in CCDSoft
  4. The background of each image is normalized to a common brightness level in Maxim-DL. The median background levels are first determined in the Information window.  The Process/Pixel Math/Subtract constant Menu is used to adjust the background pixels in each of the 3 images to a value of 50.
  5. These three normalized R,G, and B images are Color Combined in CCDSoft.  The color ratio sliders are used to correct for the differential atmospheric transmittance of light for each wavelength at the average altitude of the object above the horizon during the exposures (See Table 17.3, page 475 in The Handbook of Astronomical Image Processing.).  
  6. Astronomical image processing programs automatically perform histogram stretches on each 16-bit R,G, and B image before combining them into an 8-bit RGB image.  I prefer to create the RGB image in Photoshop where I have much more control over the histogram adjustments before they are converted to 8-bit files.  I use the RGB from color combining in CCDSoft as an internal control and refer to it as I tweak the colors in Photoshop.     
  7. Color combining using Photoshop:  The 3 normalized R,G, and B FITS images are converted to 16-bit TIFF files and imported into Adobe Photoshop 6.0.  Linear histogram stretching (black & white point adjustments) are  applied equally to each image to preserve color balance.  This process is stopped when the middle brightness levels begin to show in the nebulosity.  No non-linear stretching is done at this point.  Each image is then converted to an 8-bit file.  The three images are combined in the Channels window to create an 8-bit TIFF RGB color image.  At this point, non-linear histogram shaping using Curves can be applied to the image (not to an individual channel, however) without altering the color balance.  A Gaussian Blur can be used to smooth the noise in each of the three channels of the RGB image.  Using the CCDSoft RGB image as a reference, the colors of the Photoshop RGB  image are tuned with Color Adjust and Hue & Saturation.  The final RGB image is not cropped.

 Creating the LRGB Image:

Please refer to Robert Gendler's article, Color CCD Imaging with Luminance Layering in the July 2001 issue of Sky & Telescope, pp133-136, or to his web site for a nice tutorial on creating LRGB images using Layers in Photoshop.

    At this point the L image is an 8-bit 2184X1472 TIFF file and the RGB image is an 8-bit 1024X1024 file.  Both files are approx 3 mB in size.  Before combining the two images into an LRGB image they are aligned using Registar (Auriga Imaging).  The L image is used as the reference and the new aligned RGB image (RGB-reg) is saved.

     The L image and the RGB-reg image are opened in Photoshop with each image enlarged by the same %. The entire L image is cut and pasted onto the RGB image.  Don't forget to check No to "Do you want to save changes?" before closing the L image!  In the Layers Menu the opacity of the top layer (L) is reduced to 50% and precisely aligned (at high image magnification) over the bottom layer (RGB) using the Move tool and/or the arrow keys.  After alignment, the opacity is brought back to 100% and the LRGB image is saved a photoshop file (.PSD) with two layers.  Since the RGB and L components are in separate layers, they can be manipulated and enhanced independently.

  This process can be repeated (LLRGB): The opacity of the top layer (L) is kept at 50% and the RGB layer color saturation is increased.  The file is then flattened and used as a new, enhanced RGB layer for the next LRGB composite.  This technique enriches the color of the new LLRGB composite as well as serving to equalize the contrast between the two layers.

   When I'm satisfied with the result I flatten the image and save it as a TIFF file.  Grain Surgery for Photoshop (Visual Infinity) is used to eliminate residual noise in the image.


STL-11000:  Image processing is essentially as above.  The camera is binned 1X1 for Luminance and H-alpha exposures.  Individual exposures of 10- to 30-minutes are used (the STL-11000M is anti-blooming).  These exposures are combined in CCDSoft to yield long effective integration times to reduce the S/N.  The camera is binned 2X2 for the RGB data.  Generally 3 to 5 individual 10 minute exposures for each color filter are combined.   LRGB, LLRGB, (R)RGB or (RB)RGB, and (HA)RGB layering are carried out as described above.