 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
| PhotoKB HomeDiscussion Groups Digital Photography Digital PhotoDSLR CamerasZLR CamerasPoint & Shoot CamerasFilm Photography 35 mmLarge FormatMedium formatDarkroomFilm and LabsOther EquipmentPhoto Technique Nature PhotographyPeople PhotographyTechnique GeneralGeneral Photo Topics General TopicsAustralian PhotographyUK PhotographyDirectoryPhoto Clubs |
Photo Forum / Digital Photography / DSLR Cameras / July 2006Image noise
O.K., I forwarded these images to Canon and they recommended that I send my 20d in for inspection. The shots were taken with manual settings at f/2.8 and 1/1000 with a Canon 70-200 L IS USM at 200mm. The first three pictures don't look too bad, but the noise is readily apparent in the last one. I guess my question is, has anyone else with this camera seen this? Is it a bad sensor, or is it just a limitation (bad camera?) of the camera? Should I send it back? Here is the link to the pictures: http://home.alltel.net/tskipper > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper Hard to tell anything with such small files. >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 8 lines] > >Hard to tell anything with such small files. I think the size is sufficient to display the problem. A discerning eye can see it easily. >>>O.K., I forwarded these images to Canon and they recommended that I >>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 11 lines] > I think the size is sufficient to display the problem. A discerning > eye can see it easily. I guess I don't have a discerning eye. Greg  Signature"All my time I spent in heaven Revelries of dance and wine Waking to the sound of laughter Up I'd rise and kiss the sky" - The Mekons >>>>O.K., I forwarded these images to Canon and they recommended that I >>>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 15 lines] > >Greg I think you are correct! >>>>>O.K., I forwarded these images to Canon and they recommended that I >>>>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 17 lines] > > I think you are correct! In that case I think you are an idiot and are being a twit. But perhaps you see yourself differently, you get that I guess. >>>>>>O.K., I forwarded these images to Canon and they recommended that I >>>>>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 20 lines] >In that case I think you are an idiot and are being a twit. But perhaps you >see yourself differently, you get that I guess. Huh? No, I don't get "it". Do you have some useful input? If not, please go away. >>>>>>>O.K., I forwarded these images to Canon and they recommended that I >>>>>>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 25 lines] > Huh? No, I don't get "it". Do you have some useful input? If not, > please go away. Seems that you are not understanding here. Three people have now said that the pictures are too small to evaluate properly, I guess it is now up to you to decide what you want to do George. >>>>>>>>O.K., I forwarded these images to Canon and they recommended that I >>>>>>>>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 31 lines] > the pictures are too small to evaluate properly, I guess it is now up to > you to decide what you want to do George. Sorry, make that four! Pics a bit small to see much - they do look a bit soft. Downscaling tidies up noise a lot - I'd have to see an original. Guy > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper Hi, You can't really give a qualitative assessment of images such as these because a) they are too small and b) jpeg compression artifacts can obscure what is really going on. May I suggest you post a link to a some full resolution image samples instead. Nigel >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 15 lines] > >Nigel The full jpeg of the worst image with metadata intact is at the same link: http://home.alltel.net/tskipper > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper Yes they are same, the last one does have a pattern, that looks more like jpeg artifact than noise, too small to tell. What ISO was used ? Looks like lousy weather conditions or late evening. Was this a high ISO situation or were you trying to pull out information from a dark image, even in RAW that is asking for noise. Just remember every camera has limitations and just because you are shooting digital doesn't mean everything will came out. It does look like you have some exposure issues. But if Canon thinks you should have the camera checked I'd do it. Amazed that hummingbird wings are not even close to being stopped by 1/1000 of a second. Tom On 6/26/06 6:06 AM, tomm42 posted the following: >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 16 lines] > issues. But if Canon thinks you should have the camera checked I'd do > it. Focus is off on most of them. Very little DoF at 2.8, so you need to carefully focus manually. What ISO, what compression for jpegs? Signaturejohn mcwilliams >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 20 lines] > >Tom O.K., a larger sample of just the worst picture is now up at the same link. The image was shot in RAW at an ISO of 200. It was converted to jpeg using Photoshop cs2 with the camera raw defaults and the highest quality setting (12). The noise isn't a result of jpeg compression since it's just as visible when viewing the RAW file before compression. >>> O.K., I forwarded these images to Canon and they recommended that I >>> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 27 lines] > since it's just as visible when viewing the RAW file before > compression. Thats better but you will need to leave the exif data intact, in CS2 don't do a save for web as that strips the data, just do a save as jpg, and if you could put up one of the others the same that would be good. Cheers. >>>> O.K., I forwarded these images to Canon and they recommended that I >>>> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 33 lines] > >Cheers. The full jpeg with the metadata intact is now there. The other pictures were there just for comparison. I'm really just trying to get some input on the one that's up. Is it a sensor problem that needs to be, or can be fixed, or is it just a limitation of the camera? > The full jpeg with the metadata intact is now there. The other > pictures were there just for comparison. I'm really just trying to get > some input on the one that's up. Is it a sensor problem that needs to > be, or can be fixed, or is it just a limitation of the camera? Was the posted image this bright on the camera LCD, or did you boost the brightness with exposure, levels, curves, or other similar software tools? The noise is unusually high, unless you boosted the brightness in software. Roger >> The full jpeg with the metadata intact is now there. The other >> pictures were there just for comparison. I'm really just trying to get [quoted text clipped - 7 lines] > >Roger No, the jpeg was processed using the camera raw defaults in PhotoshopCS2. >No, the jpeg was processed using the camera raw defaults in >PhotoshopCS2. But the defaults will change exposure and other factors. What Roger is asking about is that if the was an underexposed image and in the conversion process the exposure and/or brightness were increased the noise will be more apparent. Set all the settings to zero in ACR and compare. -- Ed Ruf (Usenet2@EdwardG.Ruf.com) http://edwardgruf.com/Digital_Photography/General/index.html >O.K., a larger sample of just the worst picture is now up at the same >link. The image was shot in RAW at an ISO of 200. That's a big mistake (the ISO, not RAW). I can't even imagine shooting hummingbirds at ISO 200. 800 is the minimum, and I'd be more likely to use 1600. 1600, well-exposed, can be far less noisy than 200 under-exposed, with the 20D. There is a popular myth that high ISO settings cause noise. They really do not cause more noise, they actually add *LESS* noise, in an absolute sense, with the Canon 20D. The reason why higher-ISO images usually have more noise is that they cause faster shutter speeds and smaller apertures, and therefore allow less light to hit the sensor. If less light is hitting the sensor for some other reason, at a lower ISO, you will get more noise than if you had used the highest ISO that doesn't cause clipping for that same amount of light hitting the sensor. >It was converted to >jpeg using Photoshop cs2 with the camera raw defaults That doesn't mean anything, really, as the default with ACR3.x is auto-adjustment. >and the highest >quality setting (12). The noise isn't a result of jpeg compression >since it's just as visible when viewing the RAW file before >compression. No, it's clearly not JPEG artifacts; it is ISO 200 readout noise, complete with horizontal and vertical banding, pushed in conversion. Signature<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS@no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>>< >>O.K., a larger sample of just the worst picture is now up at the same >>link. The image was shot in RAW at an ISO of 200. [quoted text clipped - 26 lines] >No, it's clearly not JPEG artifacts; it is ISO 200 readout noise, >complete with horizontal and vertical banding, pushed in conversion. Thanks again for the input. I'm still fairly new to digital (about 8 months). I guess I'm still thinking in terms of film (use the lowest ISO possible). Apparently this doesn't apply as much with digital. I'll try the same shot with same lighting conditions at ISO 800 and 1600 and see what I get. Thanks. >>O.K., a larger sample of just the worst picture is now up at the same >>link. The image was shot in RAW at an ISO of 200. [quoted text clipped - 18 lines] >That doesn't mean anything, really, as the default with ACR3.x is >auto-adjustment. The on only acr version I have in CS2 is ACR2.4. But your right, it still uses auto-adjustment, although the adjustments can be altered at will. >>and the highest >>quality setting (12). The noise isn't a result of jpeg compression [quoted text clipped - 3 lines] >No, it's clearly not JPEG artifacts; it is ISO 200 readout noise, >complete with horizontal and vertical banding, pushed in conversion. Yes, it appears the noise is a result of user error. I took the same shot today at the same time (between 8 and 9 PM.) with a tripod and flash, and no noise. The new picture is up at the same link: http://home.alltel.net/tskipper. It's still not quite what I'm after, but looks O.K. I also took some shots of the feeder sans hummingbird at ISO 800 and 1600 respectively without the flash. At 800 some noise in the blurred background was still visible and at 1600 it was gone altogether! I'm still learning. I took a trip to Tahiti in April, took well over a thousand shots without the noise issue. Apparently it only manifests in certain low light, underexposure conditions as this was the first time I observed it. As a side note, I'm still wondering why the tech at Canon told me to send the body in for repair after seeing the too small jpeg's ( as noted by Pete D and others, who by the way seem to have lost their voices since losing things to criticize). Perhaps if I send it in, I'll get an updated, improved sensor? Anyhow, thanks for the input. >>No, it's clearly not JPEG artifacts; it is ISO 200 readout noise, >>complete with horizontal and vertical banding, pushed in conversion. >Yes, it appears the noise is a result of user error. I took the same >shot today at the same time (between 8 and 9 PM.) with a tripod and [quoted text clipped - 4 lines] >1600 respectively without the flash. At 800 some noise in the blurred >background was still visible and at 1600 it was gone altogether! That's to be expected with manual exposure, and the same Tv and Av values, or with Tv-priority-auto mode, with the lens open at all ISOs. >I'm still learning. I took a trip to Tahiti in April, took well over a >thousand shots without the noise issue. Apparently it only manifests [quoted text clipped - 4 lines] >have lost their voices since losing things to criticize). Perhaps if I >send it in, I'll get an updated, improved sensor? I doubt you'll gain anything but the lack of your camera for a few weeks. There is no evidence that you have a problem, nor do different vintage 20D or 30D cameras seem to have different RAW noise levels; Noise is determined firstly by absolute exposure (Tv and Av values for the lighting) and secondly by the ISO setting (if Tv and Av are fixed, then the lowest ISO gives the most noise; lower ISOs only give less noise when Tv and Av change to increase absolute exposure). Signature<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS@no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>>< >>O.K., a larger sample of just the worst picture is now up at the same >>link. The image was shot in RAW at an ISO of 200. > That's a big mistake (the ISO, not RAW). I can't even imagine shooting > hummingbirds at ISO 200. 800 is the minimum, and I'd be more likely to > use 1600. I don't see why. The picture (the one which is there today, at least) looks like flash usage to me, and enough flash (maybe multiple units) can put out enough power for ISO 200. I'd try for 1/250s, that being the X-sync speed of the 20D, and freeze the wings by having the flash(es) being the major or only contributing light source --- and by setting them to low power, they'll be on for maybe 1/5.000s or 1/30.000s ... > 1600, well-exposed, can be far less noisy than 200 > under-exposed, with the 20D. Think of it as a microphone , a pre-amp, and a not-too-good tape recorder. To play, you play the tape in another tape player and use a amp on the way to the loudspeakers. Now, a DSLR has a high quality microphone (the sensor), a good pre-amp (which is controlled by the ISO settings) and the tape recorder is a bit hissy and records some static. (Sorry for the tape recorder, but that's the current state of the payable art for digital-analog converters: 12 bit and some read noise.) P&S, by comparison usually have a cheap micro, often bad-to-average accustics (the optic elements), a pre-amp and a tape recorder which probably doesn't matter too much with the micro used. But they are much cheaper and much smaller. (Of course, there are exceptions.) Well enough for loud stuff (if you don't mind the accustics), not suited for silent stuff. If you record a band PLAYING VERY LOUD, you can use a very low pre-amp setting and still almost saturate the tape, so that any noise recorded is completely overwhelmed by the signal. Actually, you'll want the signal just before the peaks are too strong for the recorder (expose to the right with RAW), if you want to edit a lot later on, so you get maximum signal and minimal noise. If you record a very silent scene (very much underexposed!), you'll probably hear some noise from the micro and some from the pre-amp, but the tape noise is worst by far. You nearly cannot hear anything on playback. So you turn up the amplifier (RAW: Exposure up!) and what do you amplify? Mostly tape hiss and crackle, a bit of the near-silent signal and a bit of pre-amp/micro noise. But ... mostly noise from the tape. So next time you do that, you turn up the pre-amp. Way up. OK, what will you record? The signal, and the micro noises, both amplified in the pre-amp --- this being a good mike, it's much signal and still little noise! The pre-amp noise is there, but again, not strong compared to the signal. The tape noise is there, but as in the loud scenes, it's smothered by the signal (+ micro noise + pre-amp noise). If you play it later on, you don't need to turn up the amplifier much to feed the loudspeakers, so you don't amplify the tape noises much, so you get a good sound, only marred slightly by the micro and pre-amp noise. (Actually, it's nearly the same amount of microphone and pre-amp noise as if you hadn't used the pre_amp a lot, and relied on the amplifier for later correction, I might add. You are missing out on the tape noise, though, which was the whole point of the pre-amp.) Thus: turn up the pre-amp, if your other choice is not using the tape to the max. Yes, you'll have more noise that way, but way less than having to cope with the tape noise _as_ _well_. > There is a popular myth that high ISO settings cause noise. Of course _situations_ where ISO 1600 is well exposed are gonna be more noisy than /different/ _situations_, where ISO 100 exposes correctly. Different situations! > The reason why higher-ISO images usually > have more noise is that they cause faster shutter speeds and smaller > apertures, and therefore allow less light to hit the sensor. The reason is that less light hits the sensor and _thus_ the inherent noise in reading the sensor is amplified along with the content. -Wolfgang >>>O.K., a larger sample of just the worst picture is now up at the same >>>link. The image was shot in RAW at an ISO of 200. [quoted text clipped - 83 lines] > >-Wolfgang Things that make you go: Hmm. >> That's a big mistake (the ISO, not RAW). I can't even imagine shooting >> hummingbirds at ISO 200. 800 is the minimum, and I'd be more likely to >> use 1600. >>I don't see why. >The picture (the one which is there today, at least) looks like >flash usage to me, and enough flash (maybe multiple units) can >put out enough power for ISO 200. Sure, if you want to spook the birds in low light, and/or get a black background (or a blurry ambient exposure). And guess what? Your dark background is under-exposed, and therefore noisy; much noisier than your bird would be at ISO 1600. A bird in sharp focus hides much more noise than background bokeh can. >I'd try for 1/250s, that being the X-sync speed of the 20D, and >freeze the wings by having the flash(es) being the major or only >contributing light source --- and by setting them to low power, >they'll be on for maybe 1/5.000s or 1/30.000s ... A higher ISO allows this, also. I always use high ISOs (at least 800) for exposures of birds that are mostly flash. >> There is a popular myth that high ISO settings cause noise. >Of course _situations_ where ISO 1600 is well exposed are >gonna be more noisy than /different/ _situations_, where ISO >100 exposes correctly. Different situations! I don't choose between taking a picture at ISO X vs a different picture in a different situation at another ISO. I choose the ISO for the scene I am shooting. I am not on an assignment to get "any" picture, with as little noise as possible. If that were the case, I'd shoot a medium-saturation purplish-magenta sheet of paper bracketed from +3 to +3.7 EC at ISO 100. I have spreadsheets here of noise levels in 20D RAW files at various levels at various ISOs; these figures are always almost exactly the same on all normally-functioning 20Ds and 30Ds. In terms of StdDev/signal, the deepest shadows (due to offset-based blackframe noise) have 7x as much noise at ISO 100 as they do at ISO 1600. Instead of have 4 stops more DR at ISO 100 than 1600, we get less than 2 stops. ISO 100, 1 stop under-exposed has slightly more noise than ISO 1600 with +1 EC, in the same situation. In the same situation, the higher ISO gives the least noise, with the same aperture and shutter speed. The only danger of the higher ISO is that your signal is higher in the output range of the A2D converter and could clip, if absolute exposure remains constant. If you're going to talk about different situations, you also need to realize that low-light situations are often too low, even for ISO 1600, so much of the feedback, satistically, from the usage of high ISOs is disappointing. It doesn't help matters much that many low-light situations also have the light sources themselves within the scene, and are often incandescent, or deep green (outdoors in the woods; about the worst light color imaginable for DSLRs). >> The reason why higher-ISO images usually >> have more noise is that they cause faster shutter speeds and smaller >> apertures, and therefore allow less light to hit the sensor. >The reason is that less light hits the sensor and _thus_ the >inherent noise in reading the sensor is amplified along with >the content. ... but there is less noise, relative to absolute signal, at ISO 1600, which is the point I am trying to make. Under-exposure at a lower ISO has even more noise after amplification, and subsequent digital PP gain. Digital gain is always noisier, unless the camera has *really* bad readout. I don't know if any current DSLRs get more noise relative to absolute signal with higher ISOs, although they probably in their noise efficiency at higher ISOs, and some people report little difference with some cameras. Now, George Preddy claimed that his Sigma SD9 gave less noise with under-exposure at lower ISOs. I don't know if that is really true. Signature<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS@no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>>< >>> That's a big mistake (the ISO, not RAW). I can't even imagine shooting >>> hummingbirds at ISO 200. 800 is the minimum, and I'd be more likely to >>> use 1600. >>>I don't see why. >>The picture (the one which is there today, at least) looks like >>flash usage to me, and enough flash (maybe multiple units) can >>put out enough power for ISO 200. > Sure, if you want to spook the birds in low light, Are hummingbirds nightfliers? > and/or get a black background How you equate flash and low ISO with black backgrounds will be your secret. Use more than one flash, if you experience problems. Lighten up that background with them. It is _not_ that hard, now, is it? Or use an artifical background. With flash on it, of course. > (or a blurry ambient exposure). That only if you allow in too much ambient light for the bird's wings. A low ISO, a small aperture and/or a neutral gray filter help. A high ISO setting actually hurts. > And guess what? Your dark background is under-exposed, No, see above, thanks. >>I'd try for 1/250s, that being the X-sync speed of the 20D, and >>freeze the wings by having the flash(es) being the major or only >>contributing light source --- and by setting them to low power, >>they'll be on for maybe 1/5.000s or 1/30.000s ... > A higher ISO allows this, also. I always use high ISOs (at least 800) > for exposures of birds that are mostly flash. "Birds that are mostly flash." Now that's a novel description of these avians. But why a high ISO? If you want mostly flash, then either your flashes do not have enough power or you want to include ambient light ... > I have spreadsheets here of noise levels in 20D RAW files at various > levels at various ISOs; these figures are always almost exactly the same [quoted text clipped - 4 lines] > under-exposed has slightly more noise than ISO 1600 with +1 EC, in the > same situation. Not knowing the complete setup of your test, I can neither evaluate nor repeat nor comment on your findings, other than that they do conflict with my personal observations. > In the same situation, the higher ISO gives the least noise, with the > same aperture and shutter speed. Doesn't jibe with what I see in my pictures. Assuming that you don't push the image, that is. > The only danger of the higher ISO is > that your signal is higher in the output range of the A2D converter and > could clip, if absolute exposure remains constant. Nope. If you expose correctly for the set ISO, you get the same answer from the A2D converter (modulo read noise), independent of the actual ISO setting. If you aren't, either the exposure or the analog multiplier is off. The danger of clipping is _not_ increased. If you, however, compare the same exposure and aperture of the same subject with different ISO settings, then that is something completely different. > If you're going to talk about different situations, you also need to > realize that low-light situations are often too low, even for ISO 1600, > so much of the feedback, satistically, from the usage of high ISOs is > disappointing. I find handheld single-candlelight lit images working fairly well. >>> The reason why higher-ISO images usually >>> have more noise is that they cause faster shutter speeds and smaller >>> apertures, and therefore allow less light to hit the sensor. >>The reason is that less light hits the sensor and _thus_ the >>inherent noise in reading the sensor is amplified along with >>the content. > ... but there is less noise, relative to absolute signal, at ISO 1600, > which is the point I am trying to make. Under-exposure at a lower ISO > has even more noise after amplification, and subsequent digital PP gain. Underexposure and pushing later is something completely different from deciding between f/4, 1/60s, ISO 100 and of f/4, 1/1000s, ISO 1600. Please don't mix them up. > Digital gain is always noisier, unless the camera has *really* bad > readout. So how would it be less noisy with a terrible readout, but a fair analog amp before it? -Wolfgang >>I have spreadsheets here of noise levels in 20D RAW files at various >>levels at various ISOs; these figures are always almost exactly the same [quoted text clipped - 8 lines] > nor repeat nor comment on your findings, other than that they do > conflict with my personal observations. You can view similar data for the 1D Mark II here: Procedures for Evaluating Digital Camera Sensor Noise, Dynamic Range, and Full Well Capacities; Canon 1D Mark II Analysis http://www.clarkvision.com/imagedetail/evaluation-1d2 Look at Table 1, columns F, G: read noise and dynamic range. For example, look at ISO 800 which shows a read noise = 4.04 electrons, and dynamic range of 1640. Now look at iso 100 with a read noise of 16.6 electrons and a dynamic range of 3190. If you underexpose by 8x at iso 100, the noise relative to the image signal that you would record would be much worse than the iso 800 image. Roger >>>I have spreadsheets here of noise levels in 20D RAW files at various >>>levels at various ISOs; these figures are always almost exactly the same [quoted text clipped - 8 lines] >> nor repeat nor comment on your findings, other than that they do >> conflict with my personal observations. > You can view similar data for the 1D Mark II here: > Procedures for Evaluating Digital Camera > Sensor Noise, Dynamic Range, and Full Well Capacities; > Canon 1D Mark II Analysis > http://www.clarkvision.com/imagedetail/evaluation-1d2 > Look at Table 1, columns F, G: read noise and dynamic > range. For example, look at ISO 800 which shows a > read noise = 4.04 electrons, and dynamic range of 1640. > Now look at iso 100 with a read noise of 16.6 electrons > and a dynamic range of 3190. Let's see ... Isn't ISO 50 an 'extended' setting (i.e. using some tricks)? Table 1 seems to imply that in the Dynamic Range, too ... Re: Table 1: How about assuming correctly exposed images? I.e. adding the fact that you would let in only 1/8 of the light at ISO 800 that you would at ISO 100? Assuming you let in 20k electrons at ISO 100, 40k at ISO 50, 10K at ISO 200 and so on, so you get a *correct* exposure: electrons Read Noise R.N. * 10.000 / how much worse ISO let in (electrons) electrons let in is it (combined) 50 40.000 30.62 7.655 n/a 100 20.000 16.61 8.305 + 8.5% (8.5%) 200 10.000 8.95 8.95 + 7.8% (17%) 400 5.000 5.56 11.12 + 24% (45%) 800 2.500 4.04 16.16 + 45% (111%) 1600 1.250 3.90 31.20 + 93% (308%) 3200 625 3.93 62.88 + 102% (721%) That indicates, to me at least, for the camera tested: - you may get a bit less noise by using ISO 1600, underexposing by one step and pushing by multiplying with 2, but that's only 2% difference. (OK, and lot's more room for highlights.) - Otherwise, choosing the lowest ISO and exposing correctly is the best way, underexposure just gives more noise. - ISO 50, 100 and 200 are quite similar, noisewise, with only 17% between 50 and 200. - ISO 400 is quite a bit noiser (<50% than ISO 50), but probably still OK, anything above is > 100% noisier than ISO 50. - Low ISOs have higher read noise, but should collect so much more electrons, if correctly exposed, that the higher read noise is negated by the even higher electron count. (Note that the relationship of the numbers stay, even if you divide the electrons by some factor, say 10 or 100 or 625 ...) > If you underexpose by 8x > at iso 100, the noise relative to the image signal that > you would record would be much worse than the iso 800 > image. Obviously. Underexposing is not a good way to get less noise. -Wolfgang >Let's see ... >Obviously. Underexposing is not a good way to get less noise. So, what then exactly did you object to in my post? You seem to be agreeing now ... or maybe not ... perhaps you thnk that there is a difference between "under-exposing" in the traditional sense, and exposing "correctly" but leaving lots of headroom unused in the capture. There really is no difference; a linear capture is just a photon+noise count with no intrinsic "whitepoint" or "greypoint". The optimum exposure index (or EC setting) at any given ISO setting depends on the contrast of the scene. There are lots of situations I shoot in where I can set the EC dial to +2 and not blow any RAW highlights. Doing that at ISO 1600 gives the same aperture and shutter speed as "correctly exposing" at ISO 400, and less noise; about 5% less in the highlights, and about 60% less in the shadows. Less posterization too (although that should normally only affect extreme PP). Signature<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS@no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>>< >>Let's see ... [you removed lots of important stuff. Don't do that this way --- I prefer to not be misrepresented by judicious omissions.] >>Obviously. Underexposing is not a good way to get less noise. > So, what then exactly did you object to in my post? The claim "In terms of StdDev/signal, the deepest shadows (due to offset-based blackframe noise) have 7x as much noise at ISO 100 as they do at ISO 1600." is, IMHO, misleading. First, pictures with ISO 1600 are more noisy, for the naked eye than ISO 100. Second, Roger's data do not indicate that your claim is technical right. I believe they say the opposite. Even if you let in a single registered photon at ISO 1600, you get 3.9 electrons read noise. At ISO 100 your read noise is 16.61 electrons, but you let in 16 registered photons ... Third, urging people to use ISO 1600 without need (e.g. where ISO 400 would suffice, without underexposure and/or pushing) but for some "deepest shadow noise" --- and that's how I understand your intention --- doesn't seem sensible at all. > You seem to be agreeing now ... or maybe not ... I agree with this statement: - It's better to raise the ISO for one stop if you have one stop headroom over, than to leave the headroom unused. I disagree with this statement: - Images with the same amount of headroom in the AD-converter are less noisy at ISO 800/1600/3200 than at ISO 100. > perhaps you thnk that there is a > difference between "under-exposing" in the traditional sense, and > exposing "correctly" but leaving lots of headroom unused in the capture. There is, if you shoot JPEG. (Shooting JPEG is a completely valid decision, if it is an informed one.) If you shoot RAW, 'correctly' exposed means that the headroom is utilized as much as possible, without unwanted clipping. Underexposed thus means an exposure that leaves that amount of headroom unused, and overexposed means that you should have had that amount of extra headroom to prevent clipping. Of course, little things like the colour temperature are needing their own bit of overhead ... > There really is no difference; a linear capture is just a photon+noise > count with no intrinsic "whitepoint" or "greypoint". There is. Whitepoint is where at least one channel is at it's physical maximum, but not clipped and the others are at their respective maximums given by white balance, also not clipped. Greypoint is thus the point where a grey card would lie. Obviously they can vary > The optimum > exposure index (or EC setting) at any given ISO setting depends on the [quoted text clipped - 3 lines] > exposing" at ISO 400, and less noise; about 5% less in the highlights, > and about 60% less in the shadows. That are not equivalent exposures, your EV has shifted by 2 stops due to changing the ISO and not adapting the rest. I tried _quite_ hard in my post to make clear that I was _not_ talking about those. Anyway, I tend to find that highlights blow themselves rather easily, but then I probably shoot different things. > Less posterization too (although > that should normally only affect extreme PP). Which is because you didn't expose correctly (i.e. to the right, since you use RAW) at the ISO 400 picture. -Wolfgang >>>>I have spreadsheets here of noise levels in 20D RAW files at various >>>>levels at various ISOs; these figures are always almost exactly the same [quoted text clipped - 26 lines] > Isn't ISO 50 an 'extended' setting (i.e. using some tricks)? > Table 1 seems to imply that in the Dynamic Range, too ... Yes it is an extended range, There is no trick, but it is not standard for a reason (see below). > Re: Table 1: How about assuming correctly exposed images? I.e. > adding the fact that you would let in only 1/8 of the light at > ISO 800 that you would at ISO 100? If you read the paragraph just above table 1, it would explain the maximum signals. At iso 50, the sensor saturates at 79,900 electrons. At iso 100, maximum signal is 53,000 electrons, and at higher iso's, the maximum signal recorded is proportionally less (e.g. iso 200 max signal is 53000/2). It has nothing to do with "letting in" light. The lens lets in light. The sensor captures it. ISO is a gain setting on the output of the sensor. The sensor has only one sensitivity, that is the quantum efficiency of the device. Higher iso's simply digitize a lower range of the sensor. For example, the sensor captures and converts light (photons) at the same rate at iso 100 as at iso 1600. Just the output is digitized differently. > Assuming you let in 20k electrons at ISO 100, 40k at ISO 50, 10K > at ISO 200 and so on, so you get a *correct* exposure: [quoted text clipped - 8 lines] > 1600 1.250 3.90 31.20 + 93% (308%) > 3200 625 3.93 62.88 + 102% (721%) Your column 4 is scaled inverse of dynamic range, but that was not John's point. People continually ask should I underexpose at low iso to simulate high iso, while also giving me more headroom. Yes you do get more headroom, but you lose in other ways, especially effective dynamic range and increased noise at the low end. For example, say you got 1250 photons at iso 1600. With a read noise of 3.93 you get a dynamic range of 1250/3.93 = 426. If you then switch to iso 100 and expose the same as at iso 1600, then you still get 1250 photons, but the read noise is 16.61, so your shadow detail gets lost and the dynamic range is dropped to 1250/16.61 = 75. Further, your image appears much noisier, especially in the darker parts of the image. > That indicates, to me at least, for the camera tested: > > - you may get a bit less noise by using ISO 1600, underexposing > by one step and pushing by multiplying with 2, but that's only > 2% difference. (OK, and lot's more room for highlights.) Yes, once camera gain reaches about 1 electron/A-to-D number, you gain little by going to higher iso. For the 1D Mark II, that trade point is between iso 800 and 1600, and iso 800 is very good. I generally do not go above iso 800 on my 1DII. > - Otherwise, choosing the lowest ISO and exposing correctly is > the best way, underexposure just gives more noise. yes > - ISO 50, 100 and 200 are quite similar, noisewise, with only 17% > between 50 and 200. Not quite. ISO 50 saturates the sensor before maximum signal. That is why it is not in the standard iso settings. With read noise of 30.6 electrons at iso 50, and down to 6 at iso 200 with almost the same dynamic range, iso 200 would have better shadows than iso 50. > - ISO 400 is quite a bit noiser (<50% than ISO 50), but probably still OK, > anything above is > 100% noisier than ISO 50. Remember photon noise (the brighter parts of the image) is the square root of the number of photons collected. So doubling iso, and halving exposure time and halving the photons collected results in a loss in signal-to-noise ratio of 41% (factor = 1/square root 2 = 1/1.41). > - Low ISOs have higher read noise, but should collect so much more > electrons, if correctly exposed, that the higher read noise is > negated by the even higher electron count. Yes, unless you are trying to detect something faint. > (Note that the relationship of the numbers stay, even if you > divide the electrons by some factor, say 10 or 100 or 625 ...) [quoted text clipped - 5 lines] > > Obviously. Underexposing is not a good way to get less noise. Yes. But there are many times I'll do just that. Example, I'm at iso 100, so I'm getting plenty of signal anyway, and I'm tracking an animal in a difficult lighting situation. I'll underexpose intentionally so I do not blow the highlights. E.g.: http://www.clarkvision.com/galleries/gallery.bird/web/c01.14.2003.img_5113.egret -flight.f-600.html Roger >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 20 lines] > >Tom It's not jpeg artifacts as the same pattern is visable in the raw file. I reposted the worst picture in the full jpeg with all metadata intact. Any input would be appreciated. Thanks > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper Ok this is a long shot but I have seen similar looking effect when I have shot through a screen, I assume this was not shot through a screen? Scott >> O.K., I forwarded these images to Canon and they recommended that I >> send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 12 lines] > >Scott No, it was not shot through a screen. That's why I think there's something wrong. There just shouldn't be that much noise. > O.K., I forwarded these images to Canon and they recommended that I > send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] > Here is the link to the pictures: > http://home.alltel.net/tskipper What does the histogram from the raw file (before conversion) look like? Does it show a fair distribution, or is it all bunched up on the left? Is the posted image with the birds on extreme right the full image, or cropped, and if cropped, what area of the image are we looking at? Has contrast been boosted, and how much sharpening has been applied? On the surface, it looks to me like underexposure of a dark background that has been brightened and possibly sharpened, thus showing noise. Without access to the original raw file it's hard to say more. Considering your quoted exposure of 1/1000 at f/2.8, that converts to 1/25 at f/16, which is actually a very slow exposure at 200 ISO. My guess is, at that exposure, you were working in very low light levels, and actually still underexposed the image. Colin D. SignaturePosted via a free Usenet account from http://www.teranews.com >O.K., I forwarded these images to Canon and they recommended that I >send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 6 lines] >Here is the link to the pictures: >http://home.alltel.net/tskipper I only see one picture here. It is hard to tell if this is abnormal, because there is no way of knowing the exposure level, and how much the image is brightened in software. Noise is extremely dependent on these factors. The image could be downsampled as well. I don't like JPEGs for gauging noise. With a RAW file of out-of-focus areas of flat color, there are usually *VERY* exact amounts of measureable noise at different exposure levels. I have checked people's Canon 30D images this way, and every one I checked had the same strength and character of RAW noise, and the same as my 20D, literally to 1 or 2%. My guess is that you under-exposed at a low ISO, which is generally a huge mistake. Low ISOs do not improve image quality if you can't get a full exposure; they RUIN it. Dark green backgrounds, unfortunately, are one of the worst offenders. The exposure in the red channel is often extremely low (several stops lower than the green), and therefore noisy and posterized. Signature<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS@no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>>< >>O.K., I forwarded these images to Canon and they recommended that I >>send my 20d in for inspection. The shots were taken with manual [quoted text clipped - 8 lines] > >I only see one picture here. Yes, I took the rest down as this was the only one that really showed the problem. >It is hard to tell if this is abnormal, because there is no way of >knowing the exposure level, and how much the image is brightened in >software. Noise is extremely dependent on these factors. The image >could be downsampled as well. I don't like JPEGs for gauging noise. The metadata is there per another's suggestion. >With a RAW file of out-of-focus areas of flat color, there are usually >*VERY* exact amounts of measureable noise at different exposure levels. [quoted text clipped - 5 lines] >huge mistake. Low ISOs do not improve image quality if you can't get a >full exposure; they RUIN it. Well, this seems to be the growing consensus. Before I send the camera back, I'll try the same shot at higher ISO's >Dark green backgrounds, unfortunately, are one of the worst offenders. >The exposure in the red channel is often extremely low (several stops >lower than the green), and therefore noisy and posterized. Makes sense. Thanks for the constructive input! |
Reply to this Message |
 Sign In
 Join
 My Latest Posts My Monitored Threads My Blog My Photo Gallery My Profile My Homepage Start New Thread Enable EMail Alerts Rate this Thread
|
©2009 Advenet LLC Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.