Sky was clear all night last night so I took two data sets of Lagoon and Trifid with the in camera noise reduction on and off.
Noise reduction OFF
Canon 5DH, ISO 1000, Hutech LPR filter, 100ED with Astrophysics focal reducer so 640mm at F6.4. 15X7minutes. Raw converted to 16bit uncompressed fit monochrome. darks,bias and flats same. Stacked as Tiffs in RegiStar digitally developed with ImagesPlus Richardson-Lucy enhancement and stretched
Gone
Noise reduction ON
Same as above except only used flat calibration in 16b uncompressed, all done with 16bit Tiff Flat was also colour balanced to grey, hence the colour difference
13X7minutes
3MB http://members.optusnet.com.au/~avan...GTR_CFA_ON.jpg
No independant adjusting of colours or curves etc. Only digitally developed using same parameters. Richardson-Lucy enhancement to both at same settings and then stretched.
The size does not reflect any difference in quality. Both originals over 12MB.
Sir, I must say, bweautiful shots, now thats out of the way by far I prefer your "ON" shot, only by doing a flip process and looking at both images closely.
Yes both images are lovely, the 5D has the lowest noise image sensor of any DSLR, even the NRoff shot shows few hot pixels. The 5D has bigger than normal pixels with its full sized 35mm sensor, making it the lowest noise DSLR in the business
Scott
Hi Bert,
Very nice pictures and a fantastic wide field, Is the Noise Reduction an Internal Dark Frame removal or some other function, does it require a longer process time, and is it any improvement over doing a DF subtraction during processing, A fantastic Camera the 5D with a really BIG image chip and the larger Pixell size makes it more sensitive again, keep those great Pix. coming.
The in camera noise reduction is better than any method in images plus. For a start there is virtually no colour noise with ICNR just tiny weak 'holes' which disappear if frames are dithered or shift slightly and stacked frames are averaged as median.
I tried using all uncompressed 16bit CFA fit files for bias,flat and dark frames and of course light (image). This gave the best noise reduction even noticeably better than lossless compressed fits. There was still some amp glow. there is none with ICNR. The selection of dark frames was also critical as I had dark frames taken before and after data I tried an average of both, just before and just after. Dark frames after worked best and this also corresponded better to the temperatures. I used the scale factor option (auto box ) in all cases as it worked better.
To get the same result as ICNR unless the temperature of the camera was the same all night dark frames would have to be taken before during and after and then work out which gives the best result. You may as well use ICNR and get it close to perfect.
The other option is to keep the camera temperature constant by enclosing it in a thermostatically controlled peltier cooled 'fridge'.
You can see the amp glow and colour noise in this pic ICNR off
9X7 min ISO 1000 darks bias and flats etc in IP.
This pic 17X6min frames at ISO1600 INCR ON only used flats for correction
Both of these shots are excellent. Thanks for posting these comparisons. This is most informative.
What I would really like to see is a single frame with ICNR ON compared to a single frame with INCR OFF after calibration with a master dark and bias. Flats wouldn't be needed. This would give the best comparison of the two methods. As it is, your two original posts have different numbers of frames in the stack. This effects the final Signal to Noise ratio and so it would be better if they were single frames.
My observations on the two original posts:
Standard Deviation in both pictures is very close, with the ICNR shot showing slightly higher SD. This is interesting because, in dark frames at least, SD is a measure of the noise. That cannot be extrapolated to these frames however as the stretches seem to be different.
I calculated signal to noise ratio for a small portion of the sky background in each picture (S/N is mean/sd). This measure shows the greatest difference. ICNR ON: 6.0, ICNR OFF: 1.5 . This clearly shows a better signal to noise ratio with the ICNR ON. For interest, I applied Noiseware with the default settings to the OFF image and remeasured the same portion of the frame. S/N was then 4.99 which is very similar to the ICNR ON.
I would really like to do the same tests with frames prepared as I mentioned above. Also I acknowledge that these tests have been performed on Jpeg compressed files. Uncompressed TIFFs would be much more accurate.
Other observations:
The ON image is much softer than the OFF image
There is more colour flaring around stars in the ON image than the OFF image. I'm not sure why this should be so.
My thoughts:
ICNR seems to provide better S/N than pure darkframe subtraction. This could be verified by preparing frames as I suggest above.
ICNR appears to be doing much more than darkframe subtraction. Because it is possible to produce a similar result using post processing noise reduction software, I assume that the INCR is doing something similar.
I think that if I had the option to use ICNR, I would choose not to for the following reasons:
Imaging time is doubled. I would much rather gather 40x3min light frames and make the darks when I am packing up (or use a library) than gather 20x3min lightframes using INCR in the same time.
Control: I feel that I would have more control over the final result. I can produce similar results using noise reduction software, but I have control over how much is applied.
Of course for me all of this is moot as I own a 300D with no ICNR and so have no choice.
Hi Itchy, I will do a single frame of each and put up a full resolution tiff crop of each. The data with the ICNR ON was at at an ISO of 1600. With ICNR off at an ISO of 1000.
Would it ge better with a single frame of the lagoon data ON&OFF as they were taken on the same night under the same conditions.
I am interested to know what is going on as obviously it would be faster to not use INCR.
Bert
Hi Itchy, I will do a single frame of each and put up a full resolution tiff crop of each. The data with the ICNR ON was at at an ISO of 1600. With ICNR off at an ISO of 1000.
Would it ge better with a single frame of the lagoon data ON&OFF as they were taken on the same night under the same conditions.
I am interested to know what is going on as obviously it would be faster to not use INCR.
Bert
Hi Bert,
Same night and conditions would be better. I guess what I am trying to find out is: Which is better, INCR without dark calibration or No ICNR with dark/bias calibration. All other things either need to be controlled or ignored. So the ideal data set would look like this.
One single light frame, ICNR ON linear converted to 16bit TIFF.
One single light frame, ICNR OFF, CFA converted, calibrated with a CFA Masterdark and a CFA master bias using auto darkframe matching and then Bayer converted to 16 bit TIFF
Each frame would ideally be taken around the same time at the same ISO and exposure length.
The CFA master dark would ideally be created by averaging around 20 dark frames taken at a similar temperature and the same ISO and exposure as the light frame. The CFA Master bias would be created by average combining about 20 bias frames. Use the same ISO but use as short an exposure as possible. The master calibration frames are easy to create. Simply convert the RAW frames using CFA no white balance and combine them using average.
This procedure should produce two frames that can be directly compared. A full frame TIFF Crop of a relatively even area of the frame would be best. A corner crop could be used, provided that it is not effected by amp glow.
There you go, something to keep you busy.
Of course, only do this if you want to. No pressure!
I used ten darks and thirteen bias frames to make the masters for frame 1904. Both frames have the same conditions 7min exposure at an ISO of 1000. 1904 ICNR off, 1908 ICNR on.
I have attached a zipped Excel file with the results of my analysis.
This is what I did.
I randomly sampled 10, 99x99 sections of each image and recorded the mean and standard deviation. The signal to noise ratio is calculated as the Mean/SD (Berry & Burnell, 2000, page181).
I also applied Noiseware professional to the ICNR OFF image and examined the same samples from the image.
It can be seen from the results that the ICNR produces a higher signal to noise ratio than the calibrated image in all samples. This confirms our earlier observations.
Interestingly, when the calibrated frames have Noiseware applied with the default settings, the signal to noise ratio improves almost to the same level as the ICNR.
Although not conclusive, these results strengthen my contention that ICNR is doing something else other than simple dark frame subtraction. As noise reduction can create artifacts, I would lean towards not using the In Camera variety.
So I guess that it still somes down to your own preference. Certainly, using ICNR will speed processing by removing the need to do dark and bias calibration, and post processing noise reduction would probably not be required either. The trade off then is still the same. Time at the scope or time in processing.
Please let me know if I have done something stupid or that my thinking is wrong.
Thank you again for posting these images and sparking an interesting discussion.
Of course I am keen to hear others opinions as well.
Cheers
(ref: Berry R & Burnell J, The Handbook of Astronomical Image Processing, William Bell Inc, 2000)
