Telescope Reviews: ATM: Planetary Monorail

A friend approached me about a year ago with issues he was having with his Canon Rebel camera. He knew that I was involved with imaging as a profession and was curious why he could not get good images of planets with his new beloved camera. His interests were not in extending human knowledge of the solar system he only wanted an impressive "off world" image he could display in his studio. I had been playing around with video astronomy for many years so I though why not combine the two.

The Rebel has several disadvantages over other streaming video devices such as web cameras. Its large sensor needs to be very square to the optical axis of the telescope. The shutter and flip mirror create large amounts of vibration that needs to be accounted for. The CMOS sensor’s roughly "60% fill factor" design is less than optimal for planetary photography and would require some serious over sampling in order to produce low contrast detail. Large amounts of oversampling would require a very large image scale. A large image scale would require some sort of seeing monitoring so that the image captures could be instigated at the correct moment but could also be terminated and erased if the seeing criteria change while the exposure was being taken.

Although some have stated the Rebel does not have enough quantum efficiency to produce good contrast on Saturn. Quantum efficiency really has nothing to do with it. Full well capacity of the imaging device dictates for the most part the possible dynamic range. If the Rebel can produce contrast information in the spiral arms of galaxies that are invisible to the eye it has enough range to produce contrast information on Saturn. As long as the wavefront entering the telescope is maintained so will the MTF of the system.

Although certainly a complex problem the issues did not seem to be totally insurmountable. To solve the monitoring problem I created a flip mirror/ projection/T-ring adapter. This is comprised of a 7% beam splitter mounted between the projection eyepiece and the Rebel sensor. This beam splitter feeds images to either a Philips or Sony video camera. The video from the camera is digitized by a high-end frame grabber board and displayed on the computer monitor. The high-end frame grabber board is an Integral Technologies Flashpoint 128. This board controls both VGA and image digitization. A very nice feature of the 128 is that it has an on board RS232 port. This was very useful for the control of the monitoring camera.

Although this board was high end in its day that day has past. It does still work in my home PC. A 200 MHz Pentium 1 with a 1.2 gig hard drive. Newer PC's are too fast on initialization and query the VGA board before it is finished testing its video memory so the PC will only come up in a 16 color mode, a mode in which the video overlay is not supported.

The video stream provided by the frame grabber is monitored by a contrast function. A region of interest (ROI) is drawn on the image from the monitoring camera and examined by contrast function. This contrast function examines the pixel differential of each of the 60 fields of video per second and triggers the Rebel to take an exposure when a set threshold is met. By monitoring 60 fields per second I am also able to monitor the seeing while the exposure is taking place for exposures longer than about 1/30th of a second. If the desired contrast threshold is not maintained during the exposure then the image is erased. It may seem that this process would keep you from missing any good seeing opportunity and that may be true , however in the real world seeing opportunities that allow for slow shutter speeds are very rare. In my area of the country they seem to occur maybe twice a year or so.

Armed with this equipment our first successful attempt was the Saturn image many of you have seen. The first problem beyond the seeing was that a really good exposure of the rings made the intensity data of the ball of the planet appear too far to the left in the histogram. Since I wanted to color balance the rings to match spectrographic data found on the web I needed the rings properly exposed. Many of the Saturn images I see have the rings either overexposed or blue white in color. I was trying to preserve ring detail and allow some intensity range for a change to a warmer color tone on the rings themselves.

The only way I could see to perform this function was to take separate exposures for the ball and rings and combine them into one image. Many people compost images of different exposures. Some people even compost images from different telescopes so this idea didn't seem out of line. All though this seems like a simple task it had its own set of problems. The major problem was the resolution on the ball ended up not matching the rings since it had less sampling in poorer seeing. To get anything at all by the time the ball exposures were taken we needed to lower the contrast trigger threshold. We took fewer samples because the seeing after we took the ring-biased images seemed to be deteriorating rapidly. As it turns out I did something stupid that I didn't realize until the next day. I had inadvertently hit the "clean" button on my window heater controller. So after the ring images the heater was warming up the window as if it needed to clear off dew. After waiting so long for reasonable seeing we couldn't see throwing away the ring data. So we used what we had. Obviously we encounter problems also and we make mistakes just like other people do in the dark.

Plumbers have good plumbing and electricians have good wiring. I certainly am no different as some of the processes I produce at work spill over into my home tinkering. Since these processes would be considered the Intellectual Property (IP) of the company I work for I am unable to give precise details. Unfortunately the fine details of the removal of the Rebel shutter/ mirror slap falls into this category. There is however very excellent process description of FFT motion subtraction by John C. Russ in his book “The Image Processing Handbook” In the second edition of his book, pages 338-341, John gives a clear example of how this process is applied to aerial photography. I am sure he explains it better than I would anyway.

The motion elimination process does have limits but the level of blur produced by the Rebel shutter or mirror slap elimination is possible. The simpler the mode of vibration the simpler the frequency transform of the blur can be. The Planetary Monorail with a simple single support beam vibrates in a simple mode. Since the vibration error mode may not be the same in all images each image has to be corrected for motion before they are stacked. I have found so far that the performance of any variation of the Lucy Richardson algorithm will be enhanced with a smaller error function in the images to begin with. This has worked much better than summing all of the random motions produced by mechanical disturbances and using the Lucy Richardson algorithm variants to sort it all out.

I know that many will find this a blur removal a confusing idea, however as a simple example of motion blur subtraction I took an image of a standard resolution chart with the camera mounted by a spring. With the slow shutter speed the effects of the mirror and shutter slap are obvious. However when the frequency transform of the blur error is divided into the frequency transform of the image, the retransformation of the image will restore a blur-corrected image. This is an area that I think all Rebel owners could profit by. Again this is as long as the vibration mode that caused the blurring can be understood and is within limits. If so, shutter and mirror induced blurring can be removed. Much of the residual processing artifacting in the corrected image is reduced by image stacking. I attached the demo motion blur images to this posting to help make the end effect a little more understandable.

The Lucy- Richardson algorithm although new to many people, it has been around for more than a decade. Many others have made their own modification to the process. I myself use a personal variant. Again since my variant approaches the limit of IP I will not discuss it in detail. Many of my changes though are the result of an article I read back in 1998 written by P. Magain, F. Courbin, and S. Sohy in pages 28-31 of the ESO Messenger #88. Surprisingly it can still be found at:
http://www.eso.org/gen-fac/pubs/messenger/archive/no.88-jun97/messenger-no.88.pdf
I found the concepts introduced by these gentlemen, as both thought provoking and visionary. Their methods of limiting the inherent weaknesses of the R/L algorithm produced major increases in image detail. Perhaps they will help others who are interested.

Since we were processing images for their presentation value and not absolute scientific value we had more leeway in processing techniques. I tend to tune images to fit the eye's exceptions. Something perhaps even the scientifically inclined tend to do. Scientifically misleading? I consider it misleading only if the feature doesn't exist in the image in either the spatial or frequency domains. The edge sharpness of the Saturn image is a good example. It implies a resolution to the eye that would otherwise be almost impossible from earth. Pixel erosion is very helpful in sharpening edges. Images can be upsampled, regions of amplitude of power can be identified, and then eroded and downsampled. After a few careful cycles images can be noticeably improved. Under the right conditions the "blurring" frequencies can be found in the frequency domain and dealt with there if they have no importance elsewhere in the image. Erosion principles are also discusses in the John Russ text.

Tonal information on the rings is often lost in many images of Saturn. We tried to emphasize ring divisions by scanning for radial intensity fluctuations across the ring plane and amplifying them and using the guassian erosion principles to replicate the new intensity waveform. The trade offs are that contrast visibility may be improved but at the expense of increasing feature size. Rather than display the slightly blue/white color of the original ring images they were hue shifted to match known spectroscopic data.

By upsampling the image you can perform sub pixel processes easily. Although the erosion process works well on fully illuminated spheres it tends to really whack parts of images with terminators if not monitored very closely.

I often will split up an image into smaller regions for processing. These regions can be square, ellipsoidal, or round. Sometimes it is better to not to use the one-size fits all in applying image-processing functions to images. Functions that may really help an image in some portions may be very detrimental in other regions. These regions are processed and repasted back into the image. This process can help in many instances where you would like to increase contrast in one area without overemphasizing contrast in another. Borders and boundaries can be eliminated in the FFT frequency plot. Since the neighborhood transitions are all an exact period they are all of similar frequencies and can be eliminated.

After both the ball and the ring were processed to our satisfaction they were recombined by the same process seen today in panoramic photo stitching. This is usually performed in the HSV, or Hue, Saturation and Value color space. By transitional merging the color axis of the border a smooth transition can be accomplished.