SaratogaSkies Jim Solomon's Astropics


Latest news

December: Test shots with new scopes/mounts

Dec 21: TMB 80/480 Arrives!

Dec 3: AP1200 Arrives!

Nov 30: TMB 152/1200 Arrives!


Printer-friendly Version

Residual Pattern Noise Analysis: 350D vs 300D

This page compares the Residual Pattern Noise (RPN) of a "stock" Canon EOS Digital Rebel (300D) and a Hutech-modified Canon EOS Digital Rebel XT (350D) with Type I filter. By Residual Pattern Noise (RPN), I mean the noise resembling a checker-board pattern that remains after proper calibration with matched Darks.

In addition to the cameras, the following equipment was used in this analysis:

The intensity of the light box was adjusted to give a median ADU value of 2048 in the 350D and was kept at that same intensity when switching cameras to the 300D. Recall that the 12-bit A/D converter in both cameras gives ADU values in the range [0..4095], making 2048 the mid-point of the ADU range. The algorithm used to create the images below is as follows. For each camera, respectively, do the following, where all operations are carried out in "CFA Space" (i.e., un-Bayer-interpolated "grayscale space"):

  1. Attach camera to scope and capture Flat Lights: 9 x 5" @ ISO 800
  2. Remove camera, put body cap on, and capture Flat Darks: 9 x 5" @ ISO 800
  3. Median-combine the Flat Darks to compute a Master Dark.
  4. Calibrate the Flat Lights against the Master Dark.
  5. Normalize the calibrated Flat Lights so that each color pane has, respectively, a median value of 20000 (in signed 16-bit integer space). (These Normalized Calibrated Flat Lights are effectively "colorless".)
  6. Kappa-Sigma (κ=3, iter=1) combine the Normalized, Calibrated, Flat Lights to make a Master Flat.
  7. Use IRIS to "curve fit" a 5th-degree polynomial to the Master Flat. Call this the Synthetic Flat.
  8. Divide the Master Flat by the Synthetic Flat.
  9. Visualize at 21000/19000.
  10. Capture to clipboard and paste into Photoshop.
  11. Crop (the same) 768x512 region at 100% zoom.
  12. Convert to JPG at 100% quality.

This algorithm produces a Dark- and Flat-calibrated image that can be inspected for pattern noise. The Synthetic Flat has the virtue of being extremely smooth — i.e., essentially devoid of pattern noise and "normal" (gaussian) noise. The results are shown below. Mouse-over the respective labels above the image to have it change from 300D to 350D, etc:

[Rebel (300D)]     [Rebel XT (350D)]

Note: the mouseover effect requires JavaScript to be enabled. If you don't have JavaScript enabled, click on the links below to load the images in their own pages:


Note the random-looking character of the noise in the Rebel (300D) and the severe pattern noise of the Rebel XT (350D). I tried some variations on the above experiment as well. Specifically, I tried different exposure lengths, different ISOs, and dividing individual Dark-Sutbracted Flat Lights by the Master Flat (instead of the Synthetic Flat). In all cases, the 350D had much more Residual Pattern Noise than the 300D, and all results resembled those shown above.

Finally, remember that these are calibrated images, meaning in particular that the Residual Pattern Noise cannot be removed via typical calibration techniques for processing astrophotos. The best one can do is dither between each Light frame and hope that the registered frames will randomize the Residual Pattern Noise and make it unrecognizable.