Making DSLR flat fields

Hi Heinz, Display technology R&D is my job but I would not recommand to use electronic displaysas a "flat" ! Sorry for our business ! Their luminance is neveruniform due to various problems, this is not very visible to the users as the human vision has a log sensitivity and the brain also applies flat correction in an incredible way ! The displays light emission distribution depends of number of parameters and particularly the view angle in caseof LCD (we should have less issue with a plasma). The mapping of it, the gamma,depend ofthe luminance level... not simple at all ! Old CRT have no angleof view problems but are not uniform at all betweencenter and corners... I did try to use a very good LCDas a tool for generating grey scale for linearity measurment. After fixing number of problems, the only result I got was a map of the view angle effect ! That even if I was using a long focus lens (~400 mm ) from far but the angle variation remained unacceptable. Another issue is the Moiré effect between the display screen structureand the CFAstructure ofthe imager. Even with a strong defocusing it's difficult to reduceit and I am not sure the"vignetting" of the lens (that's theonly issue to consider, Fixed Pattern Noise is negligible for long time... ) would be the same if we are not set at theinfinity. I personally use only long focus for photometry, this provides the largest aperture by far (surface ! not F#, we are looking at stars...)A 200 or 300 mm lens gives a field of 5 ~ 7 degrees that's perfect to cover the field of interest and F/D:4gives a 50~70 mm diameter aperture that's very good for accumulating photons/electrons ! I do thevignetting modeling from exposures of a good A3 - non-glossy -photopaper illuminated with either the sun or an halogen placed at a good distance ( ~5 meters) I did check it with a photometer, it's OK (0.2% in theory...). I also defocus somewhat to reduce the remaining grain from the photopaper, even that very small onewas noticable in the noise measurment ! Anycommon paper was terrible! I agree it's more difficult for a short focus lens, but why not try a longer focus one ? You would get a better SNR. The "vignetting" of lenses is a serious problem as its level is easily several 10% in the corner of the image andsoon 5~10 % in the center area,sometimes much more.This is by far the strongest cause of instrumental error in our process. But its not easy to compensate it in good condition. One of the issue apart getting a uniform target is the strong shot noise, stacking 9 images reduces it by only a factor 3. Best is to generate a polynomial model of that vignetting but I don't think it's common in existing software (it's done in DSCs ! apparently not common in DSLR) I think we need a tool for this, one more subject in a long list... Yours truly, Roger

Hi Roger, Thanks for your in-depth reply, this saves me a lot of frustration trying to do something which others have proven to not work already :-). This idea about a synthetic flat sounds exciting... CS Heinz

Iam using an ST9 with a 50mm lense. That gives me a 10 degree field. For flat fields, I place a frosted plasticsphere over my lense as a diffuser. It's a part from a light fixture that I picked up at a local hardware store. I then aim the camera at the zenith on an overcast night when the cloud cover is uniform. I take a couple of dozen exposures in each filter. I rotate the camera on the tripod between each exposure such that I cover 360 degrees during the set of exposures. I then do a median average of each set of exposures for each filter. This has given me good flats to work with. I have repeated this proceedure ever couple of months but have found that the flats are very repeatable. Hope this helps. SAM

Hi Sam, I am OK with your idea. I was not sure a uniformilluminator at short distance was optically equivalent with a flat at infinity. I did check it, it seems ok both by experiment and simulation (Igota differenceof a couple of0.1% that's probably an experimental error) It's also ok to put aflat good (non-glossy) photopaper at some distancein front of the lens (0.5 ~1 m),illuminate it from distant sources (5~10 m), have the lens at infinity and working aperture setting. Theremaining issue is the shot noise that's annoying as the imagerflat fixed pattern noise itself is near zero on recent imagers! The flatremains mandatoryonlyfor correcting vignetting of optics, it'sa shame adding noise doing it. Yours truly, Roger

Hi Roger, As well as correcting for vignetting, flats also correct for dirt in the optics and on the sensor and variations in sensitivity from pixel to pixel. As you imply it is absolutely imperative that the flat is taken with exactly the same aperture and focus settings as used forthe lights. (and with any filters in place, if used) Shot noiseneed notlimit the final precision.Eachflat should be exposed up to the point where the brightest pixels are just still in the linear region (usually approximated by exposing to 50%) If thecounts are still not high enough to reduce the shot noise sufficently then several flats can be averaged until it is. The read noise is potentially an issue, though Isuspect it is not a limiting factor inthe final precision provided the flats are well exposed. Robin

Hi Robin, very pleased to read from you. I think this is very depending of imager being used. Fromevaluations I made on recent imagers the FPN flat part (pixel gaindispersion ) is extremely low now (I remember making cameras at atime it was at 20% level ! ) It's now onlya couple of 0.1%. When averaged on 100~200 pixelsin the measuring aperture it's just nothing. The vignetting of optics isat another magnitude, at best 20% for long focus and often 50% ormore for short focus atfull aperture ! And here averaging doesn't help... Linearity of recent (CMOS APS-C size) imagers isalso very good on the full range, I made an evaluation that shows better than 0.5% (at that point I am not sure who is worst...) that on the full range up tovery nearthe imager saturation (the worst being atcurvefoot!). By the way the best(in that case) for the shot noise is to expose atnear the full range, let say 80% for safety. The read noise is very low compared to the shot one. It's typically 4 e- today. At 80% of 30000 e- the shot is 155 e-. After averaging 25 images it remains at 31 e- and the read noise is below 1. Today the true ennemy is the shot noise in any case. In that exemple it remains at a level similarto the gain dispersionthe flat process is suppose to eliminate !For smaller format (smaller pixels), like used in DSC,it's much more sever. You are right dusts are truly annoying, but normally we should not haveso much of them in DSLR and very few with DSC. The probability of having them just on a star is not very high. Dust beingnear a pupil are no issue as they can't be imaged. The problem is at surfaces near the imager (IR filters...) and I think the best is to avoid/cleansuchdustas much as possible ! Reason is that their imaging is extremely depending of the light path and I am not so sure the flat process could be 100 % effective in that case. The last issue is to get a flat target without some gradient, if people have nophotometer I do notsee how they could put it underfull control, I always find a couple of %. Yours truly, Roger

Hi Roger, Ithink we are comming at this from slightly different directions.My experience is of using astronomical flats for photometry, and (even more challenging) spectroscopy withcooled astronomical cameras using CCDtechnology .The effects of astronomical flats and their importance has beenwellstudied based on the experiences of professional observers. Your calculationsconfirm what I saidthat provided you make sure you collect enough photons, (and provided your flat light source is bright enough that exposures can be kept short so that thermal noise does not become an issue) none of the noise sources in flatsneed have asignificant impact on the final precision. egwith say 30000 e/frame x 10 frames x 100 pixels in the photometry aperture the effect of shot noise in the flat is below the 0.02% level. On the other hand dust and dirt in optics for example has proved to be an ever present variablewhich can make a nonsense ofthe data andis the main reason why all professional astronomers take regular flats even when the optical train may not have changed. For illumination, some people are claiming good resultsusing electroluminescent panels. I am not convinced about their likely uniformity but it should be possible to verify and reduce the effects eg byrotating the panel between exposures. I haveone on order to test. Robin

Hi Robin, Anyhow in the flat itself, even pushing the averaging to 25 frames, thepresent imagers gain dispersion from pixel to pixel and residual shot noise havesimilar magnitude (from measurments made recently, no considerations coming from past technologies). On that particular view point there is no more reason to make flat... But it's clear there are other reasons to make flats, in particular the lens vignetting (several 10% ! ) andmaybe dust issues. But personally I am not convinced - in specific case of photometry -( I am in for some time also, including professionally in imaging technologies) dusts are so much a problem. This is very depending of the case: a DSC or even a DSLR is not the complex optical train ofNasmyth and Coudé configurations used by professional astronomers.The factpixel gain dispersion can be neglected opensother ways addressing theissue, this is what is interesting to me and I am experiencing on it. I have also some doubts about electroluminescent panels uniformity (as a display tech guy I have never seen any screen uniform at % level !), but if used at (or near) pupil level thereshould benoimpact of their uniformity on the uniformity of theimage at thefocus plan. Yours truly, Roger

Hello! In my humble opinion, one should expect to encounter many problems when not properly flatfielding the data. One example is following: Some years ago Ihad to test the error coming from (re)positioning filter wheel during the night. I observed open cluster h Per with 60 cm F/12 Cassegrain telescope, using two sequences which both were 50 frames long. I started observing run by taking very good S/N twilight flats in Bessel V filter and then waited for dark. Filter was not moved before the end of first series and movement of the telescope was as minimal as possible. After first series, I took second series, during which Ichanged filter before every frame as V->B->V. Thats why first series looks shorter - I used automatics to take those frames but second series was taken completely manually. All those 100 frames were carefully calibrated by substracting bias, dark frame and dividing by good flatfield frame. Measurements were carried out by IDL aperture photometry routines (basically Daophot), using adaptive aperture which takes into account the FWHM of the stellar profile. Aperture _radius_ for both stars was always 1 times average FWHM from well-exposed stars on each frame. Then I picked two normally exposed, known nonvariable stars which had similar colours, close to the center of the frame and not very close to donuts caused by the dust on filter or CCD window surface. The result is what you can see on the graph. It is interesting to note, that formal instrumental error estimate is always the same, 0.004 magnitudes. When computing additional statistics, average and standard deviation for both series, I got 1.680 and 0.005 for first, 1.687 and 0.007 for second series. The only parameter which changed before each frame during second series, was movement of the filter wheel and it's not so accurate repositioning. Altough both stars were not directly affected by dust donuts, from frame to frame changing "high frequency component of the flatfield" is enough nonuniform to give rather high scatter... Conclusion: when deviation of the differential photometry did not suffer very much (0.005 -> 0.007 magnitudes), averages are off by almost 1%! But we could expect averages much closer to 0.1% level....