Talk:Image sensor format
Do we need another page?
There's lots on this topic already in other places, but the organisation needs some improvement, and in some cases so does the phrasing and even the accuracy. And, most articles on particular camera models don't link to any of it in a particularly helpful way. And it's a hot topic... I often get asked "I've just bought a digital SLR and some lenses from the same manufacturer who made my 35mm gear, and the old (or new) lenses seem to fit the new (or old) camera but they don't work very well together" or "the lens mounts look the same and the shop said they'd be fine but they've been designed so they don't actually fit" or similar.
After a lot of wrangling, I decided that IMO there's a separate, encyclopedic topic here that's not covered by the existing crop factor article and the section at Digital photography#Sensor size and angle of view. I hope this article will expand at least to a comprehensive list of formats, similar to the film format article, with a layman-accessible but technically accurate definition of the topic.
- I'm not so sure I'm happy with the article myself, but I'm trying to at least make it correct. It might make sense to merge with crop factor in one direction or the other, and to also merge in the info from the camera article, and maybe expand to cover other effects than just the first-order angel-of-view thing. Sensor format is certainly a key parameter in digital cameras, not just for how a lens works on it. Dicklyon 04:41, 7 April 2007 (UTC)
What to name it? There doesn't seem to be a standard term. The phrase image sensor format seems mainly used by Four Thirds System vendors and enthusiasts according to Google, and describes the topic most accurately IMO. Andrewa 03:23, 7 April 2007 (UTC)
- The term sensor format as in "1/3- and 1/2-inch sensor format cameras" was in use in 1993 [1] way before the 4/3 thing. They chose to use 4/3-inch format as well as 4:3 aspect ratio, and based their name on that coincidence. Going back to 1978 [2] I find "vidicon cameras use a one inch format." The terminological trend seems compatible with the current title. Dicklyon 04:15, 7 April 2007 (UTC)
Original research
It seems to me that this page simply creates a term that isn't used widely - and isn't particularly accurate. There are a huge number of different image sensors out there - just as there are a huge number of differently sized cars. Doesn't mean there has to be a wiki page titled "car formats". Do you start to mention larger sensors used on medium format cameras, rangefinder formats, point and shoot sensors, superzooms? Of course people have trouble deciphering terms used in emerging technology, unfortunately wiki isn't the place to develop the terms. They should be already well established, with established definitions in reliable secondary sources. --Hmette 06:55, 7 April 2007 (UTC)
- A further thought... there is some coverage of this topic on the digital photography page that is probably a better place to keep it. From a quick read of that page, some of the material is becoming quite dated also and could use more input. --Hmette 06:55, 7 April 2007 (UTC)
- The table in that article has a bit of original research in it; the referenced source shows the diagonal, which is pretty standard, but the table there omits that and shows the area. And it adds the column "back" with no explanation. I made a better one, I think. Dicklyon 04:26, 9 April 2007 (UTC)
- I think that representing the size of the sensor by the area rather than the diagonal is a rather trivial calculation to be considered WP:OR. It's just presenting the same information in a different form... rephrasing mathematically rather than verbally. We rephrase things all the time without calling it original research... in fact the GFDL often obliges us to! Andrewa 14:31, 9 April 2007 (UTC)
- The table in that article has a bit of original research in it; the referenced source shows the diagonal, which is pretty standard, but the table there omits that and shows the area. And it adds the column "back" with no explanation. I made a better one, I think. Dicklyon 04:26, 9 April 2007 (UTC)
- There is indeed good material there, and there's already a link to it from this article.
- I agree that the terminology is a problem. I'll be quite happy if this article is merged, renamed or deleted, just so long as the information it contains is preserved in an accessible way. The motivation for creating it was quite simply that I wanted to refer someone to Wikipedia for information, and it wasn't easy to find it here. The information was here, but finding it was embarrassingly difficult, even for an old hand. Room for improvement. Which is what we're about, of course.
- However, the term wasn't created for or by this page, as you yourself admit above... it's not widely used (997 Google hits), but it is an existing term used for this particular topic, and I couldn't find another one. I'm very open to suggestions. Those few hits do include the official Olympus site and several others which I'd regard as reliable secondary sources.
- Or, we could expand the section at list of film formats, or just provide some better links to the information at digital photography. A refactor of some of the various stubs describing crop factor and similar terms would also be in order, and some standard links from the infobox used in articles on particular camera models are overdue. Lots of good options. Andrewa 01:46, 9 April 2007 (UTC)
- A further thought... there is some coverage of this topic on the digital photography page that is probably a better place to keep it. From a quick read of that page, some of the material is becoming quite dated also and could use more input. --Hmette 06:55, 7 April 2007 (UTC)
- I'm fascinated that you describe the term as not particularly accurate. The car analogy doesn't make a lot of sense - there's no discussion about standardisation of car formats so far as I know, and if there were then a term for these formats would probably arise, and then we could consider an article for it. Similarly, AFAIK there's no move to standardise these other aspects of camera design; If and when there is, we can and should document it.
- But yes, of course we can expand the article to include larger format sensors. That's what a stub is for! Andrewa 02:06, 9 April 2007 (UTC)
I find 42,600 google hits for "sensor format". With the redundant word "image" in there it's a lot less common, but that helps to clarify the topic, so I'd leave it. I proposed over at full-frame digital SLR that that article and crop factor ought to be merged into this one, since their topic is also all about sensor format. It's a term widely used in the industry, analogous to film format, and will help to unify a fragmented bunch of articles if we handle it right. And yes I would add the medium format sensors; not sure if any of those other things correspond to sensor formats. Sensors (and lenses to match) really are designed to a finite set of discrete formats, unlike cars. It's not O.R. Dicklyon 02:02, 9 April 2007 (UTC)
- That supports a move to sensor format, which I guess would answer some of the issues raised. But I'm not convinced there's a problem to solve. The claims that image sensor format is a new term I'm promoting are just not true. Andrewa 14:25, 9 April 2007 (UTC)
Image Sensor Format, Crop Factor, etc, all make sense to me, Ive heard/read them being used all over the net and in real life, wouldn't it be logical to state that crop factor and Digital photography#Sensor size and angle of view are the result of different Sensor Format? It might need to be stated in the Introduction of crop factor since it dose'nt explicitly do so. --58.110.131.179 (talk) 22:23, 18 November 2007 (UTC)
Video camera sensors
Shouldn't they be mentioned in here to? Kricke (talk) 00:49, 7 February 2008 (UTC)
- That'd be great :-) Do you have any authoritative information on them? ǝɹʎℲxoɯ (contrib) 05:45, 7 February 2008 (UTC)
Sensor size on cameraphones?
Anyone who knows?MaxPont (talk) 08:59, 18 March 2009 (UTC)
Scaling of S/N with size wrong?
Perhaps I'm jumping the gun here and need to go and do more research, but doesn't SNR scale linearly with sensor area, not square root of area? All other things being (roughly) equal, readout noise per pixel is constant, photon to electron conversion efficiency is equal, and noise is dominated by quantisation, so the number of photons being counted (the signal) increases linearly with area while noise remains constant. Therefore SNR increases linearly. Only if the noise floor is varying would the square root law would apply. --Adx (talk) 10:16, 24 February 2010 (UTC)
- I did jump the gun. I read the reference, turns out it is actually the original article where I learned about all this. Oops. My idea above is basically right, but what I had missed is that photon arrival is random, so the count itself becomes random (introduces noise). For some reason I was thinking it would "average out" for very large numbers (eg 1 trillion photons). Of course the noise in absolute terms (ie number of photons) adds up as you capture more of them. That's actually a fairly important point, that the bulk of the 'sensor' noise is not generated by the sensor or camera, but by the light itself. Adx (talk) 11:58, 26 February 2010 (UTC)
Bridge
I would say a bridge camera is one that bridges the gap between a compact and a DSLR. A 4/3rds camera isnt a bridge camera, its a small DSLR. In fact a micro 4/3 camera costs more than twice as much as an entry level dslr so it isnt a bridge camera at all in the sense of bridging the cost gap. A bridge camera is something like a fuji s200exr or such-like. Something that offers DSLR like features without the annoyance of interchangeable lenses etc, and at a LOWER COST. —Preceding unsigned comment added by 92.238.100.136 (talk) 02:37, 18 April 2010 (UTC)
- I'd say the "bridge" is more of a size, weight, and capabilities thing than a cost thing. See bridge camera. Dicklyon (talk) 00:32, 27 April 2010 (UTC)
The entry "Bridge camera" on Wikpedia provides a very different definition of such cameras than the one offered in this article. According to that entry, a Bridge Camera does NOT have interchangeable lenses. So I think this contradiction MUST be solved. All the more when considering that that entry is hyper-linked into this page, adding to the confusion ...!!! MarcusGR (talk) 19:05, 21 September 2011 (UTC)
Incorrect statement
The article stated "Because of their larger sensors, DSLRs can generally take high-quality pictures at ISO 1600, 3200". I suggest to remove that part. Perhaps the citation is missing because there is no credible source that would support this. Larger sensors are not inherently more sensitive. They capture more light, but the light is also spread out over a larger area so the intensity per mm^2 doesn't change. Larger sensors inherently have higher dynamic range (because any given amount of light is spread over a larger area) but they are not inherently more sensitive. The advantage of larger sensor format cameras over smaller sensor format cameras comes from the bigger lenses (remember that you have to scale f-stops with the format too, so a P&S f/2.8 lens takes in a lot less light than an FF f/2.8 lens). I took the liberty of removing the statement I dispute because the citation is missing anyhow. ClassA42 (talk) 22:46, 20 July 2010 (UTC)
I also dispute the other statements in the same section which claim that larger sensors have a better SNR. This is only true if you can increase the light with the sensor size. For any given scene and equivalent images (same FOV/DOF/shutter speed), the amount of light is the same for all sensor sizes. Hence, a larger sensor cannot capitalise on its higher dynamic range and only the latter helps to increase the SNR.
I find the current content that alludes to smaller pixels being noisier highly misleading as "pixel noise" is irrelevant. It is "image noise" that matters, i.e., one must compare noise at the same output size, not at the level of individual pixels. -> More Pixels offset noise.
- You are correct. On the other hand, people do often compare at same f-number, or at same ISO, rather than at same DOF. In that case, the larger format gives less noise.
- True. I think we agree that the assumptions made in the comparison (constant f-number vs constant DOF) should be made explicit. I'd vote for briefly mentioning what happens in the alternative scenario.ClassA42 (talk) 04:30, 23 July 2010 (UTC)
- It's hard to compare such things sensibly without starting arguments with people who only see it one way. I agree with you that the idea of smaller pixels being noisier completely misses the truth about what's going on, which is all about light, not about pixels. Dicklyon (talk) 05:51, 21 July 2010 (UTC)
- This bit you took out is not wrong, since a larger format exposed at higher ISO can still get more light to work with:
- "Because of their larger sensors, DSLRs can generally take high-quality pictures at ISO 1600, 3200[citation needed] , or even higher speeds, while compact cameras tend to produce grainy images even at ISO 400."
- I still maintain that it is wrong. At the same ISO a larger sensor is not more sensitive and cannot collect more light either (the same total amount of light is spread over a larger area). The reason why larger sensor cameras fare better with low-light situations is because they have the larger (faster) lenses in front of them. There is no other (sensor-format related) reason. ClassA42 (talk) 04:30, 23 July 2010 (UTC)
- Not quite; same ISO (with same exposure time implying same same f-number) implies same focal-plane illuminance. That's more total light, and more light per pixel.
- That's obviously not correct as one can vary the ISO independently of the total light coming in (including the case where the lens cap is on the lens). We are comparing different sensor sizes here and to make that meaningful one has to compare equivalent images. Equivalent images have the same DOF, ergo the same aperture (not f-ratio). Hence the total light is the same for both sensors.ClassA42 (talk) 00:32, 28 July 2010 (UTC)
- But you had said "at the same ISO", which implies same shutter speed and f-number, not same DOF. If you keep DOF and absolute aperture the same, and same exposure time, to get the same picture, then you have same total number of photons, and same number of photons per pixel. In that case, format size and sensor size don't have much effect. To the extent that there's a second-order effect of size, it's usually that the larger pixels, with higher capacitance, have higher read noise, including imperfectly cancelled kTC noise. So in that case, bigger is a (slight) disadvantage. Dicklyon (talk) 03:55, 28 July 2010 (UTC)
- Sorry, I don't understand. I talked about sensor sensitivity and that is only governed by the ISO setting. F-stop and exposure time will just change the total light available to the sensor but not affect its sensitivity. I agree that sensor size has no effect on sensitivity (albeit on dynamic range). I also agree with your statement about read-out noise.ClassA42 (talk) 04:43, 29 July 2010 (UTC)
- But you had said "at the same ISO", which implies same shutter speed and f-number, not same DOF. If you keep DOF and absolute aperture the same, and same exposure time, to get the same picture, then you have same total number of photons, and same number of photons per pixel. In that case, format size and sensor size don't have much effect. To the extent that there's a second-order effect of size, it's usually that the larger pixels, with higher capacitance, have higher read noise, including imperfectly cancelled kTC noise. So in that case, bigger is a (slight) disadvantage. Dicklyon (talk) 03:55, 28 July 2010 (UTC)
- That's obviously not correct as one can vary the ISO independently of the total light coming in (including the case where the lens cap is on the lens). We are comparing different sensor sizes here and to make that meaningful one has to compare equivalent images. Equivalent images have the same DOF, ergo the same aperture (not f-ratio). Hence the total light is the same for both sensors.ClassA42 (talk) 00:32, 28 July 2010 (UTC)
- Not quite; same ISO (with same exposure time implying same same f-number) implies same focal-plane illuminance. That's more total light, and more light per pixel.
- I still maintain that it is wrong. At the same ISO a larger sensor is not more sensitive and cannot collect more light either (the same total amount of light is spread over a larger area). The reason why larger sensor cameras fare better with low-light situations is because they have the larger (faster) lenses in front of them. There is no other (sensor-format related) reason. ClassA42 (talk) 04:30, 23 July 2010 (UTC)
- "Because of their larger sensors, DSLRs can generally take high-quality pictures at ISO 1600, 3200[citation needed] , or even higher speeds, while compact cameras tend to produce grainy images even at ISO 400."
- Well, ISO means different things to different people. Usually "sensitivity" is not quite among them, when you're talking digital. My point was that if you interpret ISO as the Exposure Index (EI), then it's what your light meter uses to set your shutter speed and aperture. Setting for twice the ISO gives you half the exposure. When someone says that camera A at ISO 1600 does as well as camera B at ISO 400, they mean that the image quality is comparable when camera A has 1/4 as much total exposure, measured as lux seconds in the focal plane. If camera A has 4X the format area, these are equal amounts of total light energy. Dicklyon (talk) 04:54, 29 July 2010 (UTC)
- You can say it's due to faster lens, though "faster" is usually taken to mean lower f-number, but you mean larger entrance pupil area. So yes, they fare better because they have larger lenses, but that's the same as saying that they are less noisy per pixel at same ISO. Dicklyon (talk) 06:01, 23 July 2010 (UTC)
- I'm fine with someone stating that noise per pixel is reduced by using a larger (faster) lens, but that statement is true independent of the sensor format! The only advantage of a bigger sensor regarding low light is that typically lenses for it exist (e.g., 50/1.4 for a FF) that have no equivalent for the smaller sensor (e.g. 33/0.9). I argue that lens availability should be treated as secondary topic when discussing sensor sizes in order to avoid the potential confusion .ClassA42 (talk) 00:32, 28 July 2010 (UTC)
- Sort of true, except that the f-number limitation is really a first order effect on avaiable light collection. Aperture area, in the fast-lens limit, is proportional to sensor area. Dicklyon (talk) 03:55, 28 July 2010 (UTC)
- I'm fine with someone stating that noise per pixel is reduced by using a larger (faster) lens, but that statement is true independent of the sensor format! The only advantage of a bigger sensor regarding low light is that typically lenses for it exist (e.g., 50/1.4 for a FF) that have no equivalent for the smaller sensor (e.g. 33/0.9). I argue that lens availability should be treated as secondary topic when discussing sensor sizes in order to avoid the potential confusion .ClassA42 (talk) 00:32, 28 July 2010 (UTC)
- You can say it's due to faster lens, though "faster" is usually taken to mean lower f-number, but you mean larger entrance pupil area. So yes, they fare better because they have larger lenses, but that's the same as saying that they are less noisy per pixel at same ISO. Dicklyon (talk) 06:01, 23 July 2010 (UTC)
- If the format area is 8X or more different, it's probably true. And this sentence is not necessarily wrong either:
- This problem is exacerbated by pixel count; doubling the number of pixels on the sensor means that each pixel is half the size, and hence noisier.
- Here, the pixels do get less light each, so a lower SNR.
- Yes, lower SNR, but only per pixel. As there are more pixels, they average out to the same noise. This is what the article I cited is about. It doesn't make sense to compare at the pixel level. Otherwise, one could just as well argue that a 2MP camera yields sharper images than a 20MP camera. Surely, the same scene viewed at 100% will look a lot sharper with the 2MP image (one will see less of the scene, but the same is true for the "noise" argument).ClassA42 (talk) 04:30, 23 July 2010 (UTC)
- Probably also less sensitivity, and sometimes more absolute noise, but it's hard to say that's a general case. To conclude that lower pixel SNR makes the image worse is, as you note, not necessarily correct. It only exacerbates the problem if the SNR is disproportionately worse, which is often the case, but not always. Finding reliable sources on this stuff is not going to provide much improvement, since most of them get is wrong, or partially wrong, too.Dicklyon (talk) 06:00, 21 July 2010 (UTC)
- If we cannot reach a consensus then I guess contentious statements without sources should be removed. I believe that would be the spirit of Wikipedia.ClassA42 (talk) 04:30, 23 July 2010 (UTC)
- Sure. You removed them, and that's OK. If someone wants to put back a version that's carefully phrased and true, it should come with citations to a credible source (that's a stronger requirement than the usual WP:RS, I think, since many sources that are published in reputable places by reputable authors are far from credible, and flat wrong often, on this topic). Dicklyon (talk) 06:01, 23 July 2010 (UTC)
- Your statement about errors in many sources is, unfortunately, very true. Even the reputable authors often slip in some incorrect statement.ClassA42 (talk) 00:32, 28 July 2010 (UTC)
- Sure. You removed them, and that's OK. If someone wants to put back a version that's carefully phrased and true, it should come with citations to a credible source (that's a stronger requirement than the usual WP:RS, I think, since many sources that are published in reputable places by reputable authors are far from credible, and flat wrong often, on this topic). Dicklyon (talk) 06:01, 23 July 2010 (UTC)
- If we cannot reach a consensus then I guess contentious statements without sources should be removed. I believe that would be the spirit of Wikipedia.ClassA42 (talk) 04:30, 23 July 2010 (UTC)
Sensor size
I have deleted the incorrect statements about noise advantages. I appreciate that the phrasing of the new text could be improved and perhaps it could reference more keyword definitions. I'd also be happy if someone felt strongly about the lack of a citation of the new statements I made and thus removed the new content. However, I kindly ask everyone to *not* put back the old incorrect noise assertions. The main reason is that they were incorrect, another good reason is that they were mainly related to pixel-density. When sensor size is increased one has (among others) the options of keeping pixel-density or pixel-count the same. Even if one incorrectly thinks that there is a negative impact of pixel-density on noise then that's a different discussion and should be linked to rather than discussed in this article.ClassA42 (talk) 01:09, 28 July 2010 (UTC)
Addendum: It would probably make sense to add another section to the article (which is referenced from "Sensor Size") that goes a bit into depth about large sensors and SNR advantages. Note that even a reputable source (http://dxomark.com/index.php/eng/Our-publications/DxOMark-Insights/SNR-evolution-over-time/SNR-and-image-quality-evolution) shows FF sensors to have a 3dB SNR advantage over APS-C sensors (see the first Fig. on p.3). This isn't wrong as long as one discloses that the comparison is done with the same lens and the same settings (in particular, same f-ratio) *and* that this will result in largely different images. The only reasonable way to compare sensor sizes is to base the comparison on equivalent images (same FOV, same DOF, same motion blur) and then the 3dB advantage (of course) disappears. Anyone in favour of adding a section with such a discussion? I could provide the first draft, if necessary.ClassA42 (talk) 03:23, 28 July 2010 (UTC)
- I disagree that there's only one reasonable way to compare. I agree that's a good way, but certainly not the only. The key, often overlooked in the literature, is to simply make very clear what the comparison conditions are. I think anything we can find in sources that is clear, explicit, and correct can be used as a basis for adding something more on this to the article. Dicklyon (talk) 03:58, 28 July 2010 (UTC)
- I agree. There are other ways of comparing (less valid AFAIC, but nevertheless applicable). The key, as you mention, is to state the comparison conditions. I was very tempted to add these to the material I've added but then thought an article about sensor size should not be a discussion about "equivalent images" or other ways of comparing sensor size. That's why I asked whether there would be support for adding a section that goes into more depth about the comparison conditions assumed. ClassA42 (talk) 04:36, 29 July 2010 (UTC)
- The following is false: "*and* that this will result in largely different images. The only reasonable way to compare sensor sizes is to base the comparison on equivalent images (same FOV, same DOF, same motion blur) and then the 3dB advantage (of course) disappears."
- I agree. There are other ways of comparing (less valid AFAIC, but nevertheless applicable). The key, as you mention, is to state the comparison conditions. I was very tempted to add these to the material I've added but then thought an article about sensor size should not be a discussion about "equivalent images" or other ways of comparing sensor size. That's why I asked whether there would be support for adding a section that goes into more depth about the comparison conditions assumed. ClassA42 (talk) 04:36, 29 July 2010 (UTC)
For example: To get the same FOV on a 1.6 Crop camera a lens needs to be 1/1.6 x focal length of a full frame camera lens. Therefor a 50mm lens on the crop has the same FOV as a 80mm on a full frame. These two cameras will then give the same composition of a scene shot from the same spot. However, to achieve the same DOF, the Crop camera would need an aperture approximately 1.3 stops larger than the full frame setup. An aperture of 4.5 on a full frame yields the same DOF as f 2.8 on the crop camera. Now the crop sensor is receiving more than twice the density of light at the sensor, so of course the noise will be reduced, however that is cheating.
The aperture of the lens determines the light density. This light is not spread out thinner, as claimed on the main page, when it is projected onto a full frame sensor. Both sensors are the same distance from the rear lens elements, whether on a crop or full frame camera. The crop sensor merely ignores more than half of the light available from the lens. Designers of lenses dedicated to crop sensors often take advantage of the smaller projection area requirement by making the lens smaller and lighter, and providing the same large aperture as a full frame lens.
But I repeat: an f 2.8 lens puts exactly the same light density on a full frame and a crop sensor. There is no dilution and the page refered to below is misleading. — Preceding unsigned comment added by Etienne B (talk • contribs) 05:56, 10 December 2010 (UTC)
- You are right that a given f-number results in the same focal-plane illuminance (which is probably what you meant by "light density"). The article is not talking about a given f-number, however, but a given aperture diameter, for the case of a given angle of view; the same condition you call "cheating" for the equal DOF. Not clear why you call it that. If you want a given angle of view and DOF, you need to select that aperture diameter, and when you do, you get the same total amount of light on the sensor (but not the same illuminance). Dicklyon (talk) 07:25, 10 December 2010 (UTC)
- an f stop is exactly related to illuminance, not the total amount of light. If it were not so, we photographers would not be able to double the shutter speed when opening the aperture by one stop, ie f2.8 to f 2.0, or when increasing the ISO from 200 to 400. In fact use the same f, ISO, and aperture on a crop and on a full frame and you will get the same exposure. This is quite elementary in photography. Etienne B (talk) 14:23, 10 December 2010 (UTC)
- Yes, that's what I agreed with. Dicklyon (talk) 17:20, 10 December 2010 (UTC)
I believe the following description on the page http://en.wikipedia.org/wiki/Image_sensor_size is false: "However, the larger area does not imply a better low-light sensitivity or higher signal-to-noise ratio. While the light collecting area is larger, this is canceled out by the fact that the same total light from the scene is spread out over a larger area, that is, the light flux per square mm is lower, for a fixed aperture diameter. Larger sensors can still be advantageous for low-light applications because typically lenses for them exist with larger apertures than are possible for smaller sensors."
The following was my attempt to correct this information:
"A given lens will provide the same density of light on any sensor at a given aperture. For sensors with the same number of megapixels (MP), a large sensor will have proportionately larger pixels. The pixels on a 10 MP Full-Frame sensor will be more than twice as large as the pixels on an APS-C (2.3 times for Nikon, and 2.6 times for Canon). The larger pixel size is one of the reasons that full frame sensors typically have both superior sensitivity and reduced noise in low light in comparison to smaller sensors. The smaller sensor is equivalent to a cropped section of the larger sensor but with smaller pixels, therefore much of the light from a full frame lens will fall off the edges of the smaller sensor and is "wasted." Theoretically, a 10 MP (Canon) APS-C sensor should perform the same as a 26.3 MP full frame sensor, all other factors being equal.
Although a given lens will produce the same depth of field (DOF) regardless of the sensor, in practice DOF is also affected by the sensor size. This is because at a given focal length the photographer must stand further away from the subject in order to frame the same shot when using a camera with a smaller sensor. Since the DOF increases with the focal distance (distance between the lens and the subject), a photograph taken with an APS-C camera will have a deeper DOF than the identical photograph taken with a full frame camera using the same lens and aperture setting. The photographer can alternatively select a wider angle lens for the APS-C camera and stand at the same focal distance as with the full frame camera to frame the same shot, however the DOF will be affected in the exact same way since DOF also increases with wider angle (shorter focal length) lenses."
However it was deleted with no counter argument or explanation of the facts. I believe my description was clear, but I will leave the false information in place as I see no way to make the explanation any clearer than this.
-Etienne B — Preceding unsigned comment added by Etienne B (talk • contribs) 05:34, 10 December 2010 (UTC)
- You replaced the explicit condition "for a fixed aperture diameter" with the ambiguous "at a given aperture", for which your revised statement would only be correct if it meant the same f-number (assuming in both cases a lens with the same angle of view). Your next paragraph starts with the counter-factual "Although a given lens will produce the same depth of field (DOF) regardless of the sensor"; it is well known the the conventional circle of confusion criterion that determines the DOF is a function of the sensor size. If you think clarification is needed, or that the info in place is incorrect, consider these points first, then try again. Keep in mind that this info has been carefully considered by some good editors for several years; study the refs on depth of field, and cite those or other refs if you want to make a substantive change. Dicklyon (talk) 07:19, 10 December 2010 (UTC)
There is plenty of confusion with respect to DOF, aperture and sensor size. I'm sorry, but I am content to leave the false information standing. Your good editors are mistaken. If you care to look at calculations for examples you can use http://www.dofmaster.com/dofjs.html. As an example Canon 50D (crop) with 50mm lens at f 2.8 from 10 feet gives a depth of field of 1.29 ft, while a Canon 5DII (full frame) with 80 mm lens at f2.8 from 10 feet gives DOF of 0.79 ft. I have used both of these Cameras, and many others, in this same conditions, and the DOF is dramatically different, as shown in the calculations. Etienne B (talk) 14:23, 10 December 2010 (UTC)
- Yes, that DOF calculation is correct. But try one with the same aperture diameters (50/2.8 and 80/4.5 = 18 mm diameter) and you'll get the same DOF. DOF is unaffected by format under the conditions stated, but not under the conditions you are comparing. Read the depth of field article. Dicklyon (talk) 17:20, 10 December 2010 (UTC)
Sensor size again
In this diff, I started to back off from some of MarcusGR's recent additions, as I believe they are wrong in many cases, and are generally unsourced, too. He is trying to rationalize why bigger formats use bigger lenses, but saying things like that the angle problem is worse with big sensors is sometimes backwards. Let's discuss the particulars here. Dicklyon (talk) 17:39, 5 November 2011 (UTC)
2/3" Sensor size: Wrong Crop Factor?
I think that the crop factor reported for 2/3" sensors is wrong, it should be 3.9x, not 4.9x. — Preceding unsigned comment added by 82.58.197.187 (talk) 19:00, 23 November 2011 (UTC)
Old fashioned style
Hi, i´m the IP which calculated the stops of the table.
General
The calculation was done because this are real sensors. Photographers mainly calculate in stops, - and this makes sense - , not in factors (new exception: crop-factor).
The sense of the sentence added is to put all witness related to sensor area into one sentence. Like it was done in old encyclopedias. I love it.
This should give a simple photographic rule like the other nice rules for photographers. The witness is, to focus on the important things, and not mention the whole universe, which is absolutely and truly related. Focus. For those who want to explain the whole universe, the nobel prize is waiting for the Theory of everything.
Statements
For a given lens speed or aperture (and the witness is, that focal length, angle of view is not needed), the image area (digital or film) is proportional to the:
- resolution. Exposure time not needed to mention: The whole universe is, was and will be radiating till....
- amount of light collected/noise/dynamic. Also light is needed, intelligent people will know this, and more: if i mention exposure time, i have to mention brightness, and vignetting and... . Focus.
- simplified theory (models), NOT the practice, especially in 4D. The references (remember, Wikipedia loves references) are as commercial as nearly all references in the whole Wikipedia, give details and are helpful.
These 3 statements are relevant and directly related.
- @Dicklyon (by the way, i read a little in your patents with greatest respect, and have a question: The images of the SD1 turn nearly black+white at high ISOs, is there a nonlinearity in color sensitivity in the Foveon sensors?): Take it as a compliment: I am quite sure you can do much better than this:[3]. Your text:
- "The listed sensors span more than a factor of 1000 in area; the area..."
- If you mean sensor areas, mention it. And as previously noted, this table lists real sensors used by photographs which calculate in stops, and additionally 1024 will give more info: binary relation.
- "the area of the sensor is proportional to the area of the aperture, in the case of fixed angle of view and f-number"
- Ooops, did you really think? This means, a sensor is proportional to itself, which is not totally wrong, but senseless.
- "and so also proportional to the amount of light that may be collected from the scene in a given exposure time"
- In order to explain THIS rule, it is not necessary but EVIDENT. Focus, remember: the whole universe is related. If someone thinks, its not evident, the light from a GIVEN scene at a given exposure time can be listed as a note (not as a reference, style like List of Nikon F-mount lenses with integrated autofocus motors).
- "and typically correlating with noise and dynamic range."
- Wrong. Especially if you think 4D, which is the minimum of the universe. See comparisons/reviews.
Finally
I used bold text in main statements to improve readability and emphasize results. At anyone who does not understand: PLEASE, PLEASE do not revert, try harder, or respect those who understand. You can ask, but all answers are already written down in this and related Wikipedia articles. Thank you very much. 77.186.106.154 (talk) 23:03, 26 November 2011 (UTC)
Response
That's a lot of text and headings to back up an edit. Thanks for recognizing that I might have a clue (but not about the SD1, which is way past my time there). I'm not sure from your comments above just what you think is wrong with my version. In your version, I can tell you that I didn't like expressing sensor size ratio in f-stops, because that's never done, and it's not immediately apparent that the ratio in stops should be the log2 of the area ratio until or unless you explain that the aperture area (mutatis mutandis) is proportional to the sensor area. And giving the exact number 1024 to explain "greater than 10 stops" is silly. Saying sensor size is proportional to image resolution is weird; I suppose you mean number of pixels, for a given pixel size; but that's only one aspect of resolution, and not the most important one, in most cases. Introducing image noise out of the blue as you did is also weird. I tried to see comparisons at the cited refs, but didn't immediately find anything relevant. And now you've added comments about how carefully all the words are chosen; if every WP editor did that for their babies, we'd be swimming in a sea of comments. I'll wait a while and see if anyone else has opinion about how to improve your writing. Dicklyon (talk) 23:49, 26 November 2011 (UTC)
- I've never heard of sensor area described in anything other than mm^2 and only ever heard of stops being used for measuring aperture sizes relative to focal length (or T-stops, the light transmission ratios for cine lenses compared to an ideal lens). The stops link doesn't mention sensor area at all (only aperture pupil area), so describing the sensors in terms of stops doesn't make sense to me. -- Autopilot (talk) 00:16, 27 November 2011 (UTC)
- Hi, please, my text was NOT a personal attack. I thought i made it very clear whats wrong or irrelevant with your text, which is mainly not a version of mine, but other statements. And:
- "giving the exact number 1024 to explain "greater than 10 stops" is silly": Calm down. Reasons?
- "Saying sensor size is proportional to image resolution is weird; I suppose you mean number of pixels, for a given pixel size; but that's only one aspect of resolution, and not the most important one, in most cases." I think you are on the wrong way, see Image resolution
- "Introducing image noise out of the blue as you did is also weird." Explain.
- "I tried to see comparisons at the cited refs, but didn't immediately find anything relevant." Try harder.
- @autopilot: if you want to calculate noise, dynamic range, exposure... it makes is easier. Uses: [4][5][6][7]. But to mention a factor and not the stops is not the main thing, although it is worse in understandability. 77.186.80.223 (talk) 15:11, 27 November 2011 (UTC)
- In the first link I was unable to find anything relating sensor area to stops. The second link and the fourth link discusses the effect of pixel pitch on dynamic range, but nothing that I saw related the range to sensor area. Additionally, in the table of "Digital Camera Sensor Performance Data", two cameras with identical sensor areas (the Canon 20D and 20Da) have different dynamic range stops listed. The third link is a bunch of forum arguments, not a reliable source from what I could tell. -- Autopilot (talk) 15:35, 27 November 2011 (UTC)
- When I said "is weird" I meant I didn't understand why you'd say that; ball is in your court to explain. And we both didn't find support in the refs, so that's in your court, too. I've reverted your changes until we can settle on something sourced to say. Dicklyon (talk) 15:54, 27 November 2011 (UTC)
Edited the DR and SNR discussion
hopefully to clarify and explain the relationships between sensor and pixel sizes and DR and SNR. Note that DR and SNR, used somewhat interchangeably, are not the same thing. I hope the new text brings that out and explains the differences, and also explains the factor of design in such things as saturation capacity. Bobn2 (talk) 00:37, 9 December 2011 (UTC)
- Sorry i had to revert it: Pixel area gives SNR and DR. Lower pixel area gives less light. Thats it.
- PRACTICAL use relates more to sensor size and sensor technology. Thats probably what you mean.
- I too think this page could gain from a SHORT, EASY introduction which sums it all up. Or better:
- A fact paragraph which lists relations probably like this
- But i mostly liked your text. Sorry for the revert. 77.186.88.115 (talk) 15:28, 9 December 2011 (UTC)
- This was the first expansion of the section that was pretty much correct, I think. It suffers from being too long and detailed, and mostly unsourced though. I was not aware of that latest 2010 Aptina white paper, and learned a few things that I wasn't aware of on how conversion gain typically scales and can be adjusted in modern sensors (since I got out of doing this stuff about 6 years ago). It's clear that Bobn knows what he's talking about. The problem is just how to make it into a good article. By the way, much of this approach (to the tradeoff between pixel SNR and resolution) was worked out long ago by Albert Rose; some of his papers can be cited. Dicklyon (talk) 16:14, 9 December 2011 (UTC)
- Bobn2 is good. Its a pity he is partly "wrong" or better: not exactly true. First solutions to shorten and improve:
- Which bit is partly "wrong"? As far as I know, everything I wrote was 'right', and what was there before, and is now, was 'wrong'. That was why I edited it. Your opening statement in this part of the discussion, "Lower pixel area gives less light (energy)added77.185.22.15 (talk). Thats it." is simply wrong,[citation needed] so if you start the discussion from that point of view, it will be wrong. The point is this, as I said in my rewrite, you cannot simply extrapolate the property of a pixel to that of the sensor, you have to understand the fact that noise aggregates differently to signal, and that both DR and SNR (which are BTW different things) are frequency dependent phenomena, so judging sensor on the basis of different bandwidths (which is what comparing the SNR and DR of different sized pixels amounts to) makes no sense. Bobn2 (talk) 23:52, 11 December 2011 (UTC)
- Bobn2 is good. Its a pity he is partly "wrong" or better: not exactly true. First solutions to shorten and improve:
- Sensor size vs. lens size and depth of field: Should link directly to the main article DOP, with a few sentences directly summing up relations to sensor format
- Sensor size and telecentricity: The considerations are IMPOSSIBLE to proof and cannot be seen on practice, should also link directly to the main article
- I was going to have a go at that section as well. Again, what is said there doesn't add up either, and has some factual problems. For a start, telecentricity is not to do with the position of the rear elements, as is claimed, it is to do with the position of the exit pupil, and typically somewhat forward of the rear element position. For a second, there is no 'increased telecentricity sensitivity of smaller sensors. This is a simple geometry of angles thing, if the system scales, there is no net change in the angle of incidence. The only aspect where this is remotely true is the colour crosstalk issue, but this is in practice not a consideration, in microlensed sensors, off axis rays simply don't make it to the silicon to cause crosstalk, all you get is vignetting. Really, corner vignetting is the only thing that the telecentricity argument applies to, and it is scale independent. Best just to take out that section altogether. Bobn2 (talk) 23:52, 11 December 2011 (UTC)
- I am thinking and preparing something. 77.186.105.60 (talk) 10:36, 10 December 2011 (UTC)
- Can you be explicit about what you think was wrong, or not exactly true? Dicklyon (talk) 18:17, 10 December 2011 (UTC)
- Do you ever read my text? Already stated 1 key-point.
- By the way: Is digital image sensor noise mainly caused by shot noise at short exposure times, and increasing noise caused by leakage current at long exposure times? Or is amplifier (thermal) noise the main limit? Thanks. 77.186.105.60 (talk) 18:35, 10 December 2011 (UTC)
- Perhaps I missed something you wrote. Do you refer to "Lower pixel area gives less light. Thats it." or something else? Because that particular point was covered in detail in Bobn2's section rewrite. Or do you refer to the "impossible to prove" bit? That's not the same as saying it's wrong, so I'm having a hard time following your points. That's why I asked you to be explicit, which is the opposite of saying "I already told you". Dicklyon (talk) 22:56, 10 December 2011 (UTC)
- I'm a bit perplexed that if you don't know what are the major noise contributions, you feel yourself qualified to make a statement like "Lower pixel area gives less light. Thats it.". At normal exposure times the major noises are photon shot noise and read noise (the electronic noises, mainly transistor noise and quantisation noise), as well as PRNU, which isn't really a noise. Leakage current is pretty much out of the picture until exposures of a second or longer.Bobn2 (talk) 23:52, 11 December 2011 (UTC)
- Can you be explicit about what you think was wrong, or not exactly true? Dicklyon (talk) 18:17, 10 December 2011 (UTC)
- I am thinking and preparing something. 77.186.105.60 (talk) 10:36, 10 December 2011 (UTC)
- This was the first expansion of the section that was pretty much correct, I think. It suffers from being too long and detailed, and mostly unsourced though. I was not aware of that latest 2010 Aptina white paper, and learned a few things that I wasn't aware of on how conversion gain typically scales and can be adjusted in modern sensors (since I got out of doing this stuff about 6 years ago). It's clear that Bobn knows what he's talking about. The problem is just how to make it into a good article. By the way, much of this approach (to the tradeoff between pixel SNR and resolution) was worked out long ago by Albert Rose; some of his papers can be cited. Dicklyon (talk) 16:14, 9 December 2011 (UTC)
Inserted links to main articles to prepare for shorting. "Bridging the gap in sensor sizes" is another candidate for shorting: "Once there was a "gap", and it was filled mainly with Nikon1 (2/3 sensors were often).
@Bobn2 & @Dicklyon: I just wanted to have any practical examples of sensors and expressed it vague, as theory is ok, (I examined a lecture in Technical Optics, exercised optics construction, have studied electronics, microelectronics, measurement and sensor technology and have 4 decades photographic experience) but i never designed an image sensor.
- A CV is not evidence of being correct.
My few measurements have shown unexpected behavior especially in SNR related to SHOT NOISE and Detective quantum efficiency for a Nikon D300 and Kodak 14nx. If you have any figures for any image sensor, i will be glad to have this info. Thanks in advance.
@77.185.22.15. Your statement that 'theory is OK' is very wide of the mark. The basic mistake you are making is treating SNR and DR as bandwidth independent, and they never are.
Your error is to confuse pixel-area, sensor area and a scaled image.
- There is no error and no confusion. your errors are:
- i) to believe that the SNR or DR in a single pixel predicts the quality of an image.
- ii) to compare the 'signal', 'noise' and 'DR' over different spatial periods (i.e different bandwidths) simply does not make sense. You will notice that in any other field, these measurements always specify the bandwidth - otherwise it would be possible to achieve arbitrarily better SNR just by band limiting, which of course is not useful.
- iii) Conflation of SNR and DR. Particularly in photographic usage, and as plotted by, for instance, DxO, these are different things. The DR is conventionally the maximum signal divided by the noise floor, and shot noise has little or no influence (depending if you take the 'noise floor' as the read noise or some minimum SNR. The SNR, when shot noise is a component, is signal level dependent, so needs to be specified at some level (as DxO does in its graphs).Bobn2 (talk) 15:04, 13 December 2011 (UTC)
Facts
For a given uniform (lux) light source, a given lens aperture and a given exposure time,
Lumen and light energy is proportional to the pixel area and image sensor area
Exactly. Your text: "Lower pixel area gives less light (energy)added77.185.22.15 (talk). Thats it." is simply wrong,[citation needed]
Summarized: "Lower pixel area gives less light!" OK? 77.185.22.15 (talk) 04:20, 13 December 2011 (UTC)
Pixels should be independent and unrelated in an image sensor (minimized crosstalk)
Crosstalk is seen as an error and designers succeeded in minimizing it around or even below noise level (Not the Foveon???). The fact that pixels are somehow arranged does not make them dependent. Just a bunch of signals, every with the same potential SNR. There is NO sensor noise, just an average of independent pixel noise.
Scaling different image sensors to the same image size relates pixels due to interpolation/average
Thats what you written: In general a larger size sensor allows the capture of images (scaled to an specific image size or pixel count) THIS MAKES THE SENTENCE RIGHT! with greater signal to noise ratio and dynamic range than a smaller sensor.
But image noise is a different article!!! 77.185.22.15 (talk) 15:36, 12 December 2011 (UTC)
Problems with facts
As Colbert notes, "Facts have a liberal bias". So let's be careful. Starting with "For a given uniform (lux) light source, a given lens aperture and a given exposure time" leaves the initial conditions vague and inadequate. From the conclusion "Lumen and light energy is proportional to the pixel area and image sensor area" we can deduce that you are referring to a constant relative aperture, or f-number; but then you don't say if you're letting the focal length scale with the sensor size or not, so we don't know if the actual aperture, or total amount of light being captured from the scene, is varying with format or not. This matters a lot in some deductions; assuming constant focal length is not compatible with letting sensor size change, and assuming constant f-number is sort of a cheat, as it lets you use more light, at the cost of reduced DoF. One needs to at least be clear what assumptions one is comparing on. If you assume constant absolute aperture, as is appropriate in a system with DoF constraint, then the noise advantage will go to the smaller sensor, for a given number of pixels.
You claim that "Pixels should be independent and unrelated in an image sensor". That's a simplification of reality, since pixels in real images tend to fairly highly correlated. Nevertheless, we can live with it. More importantly, you need to assume that pixel noises are independent to do the kind of noise analysis that Bobn2 did; he had that right. And he included the crucial noise source read noise, a noise that doesn't depend on light level, that is needed to understand how noise and resolution trade when subdividing pixels; his analysis was over-simplified perhaps, in assuming a particular dependence of read noise on pixel size, but not bad.
I'm not following who said what or what fact you're proposing in the your last bullet point.
The problem, as I said before, with Bobn2's rewrite was it was too long and detailed and unsourced. Maybe we can move toward its correctness improvement in smaller steps, citing sources along the way, and end up in a better place. Dicklyon (talk) 16:00, 12 December 2011 (UTC)
- My comments above were mangled into the mess below, so I have re-inserted them above – Dicklyon (talk) 05:39, 13 December 2011 (UTC)
As Colbert (not my model)77.185.22.15 (talk) notes, "Facts have a liberal bias". So let's be careful. (i am, not the tiniest error in all of MY text above. Or else i am really thankful if anybody shows ANY error or even incompleteness.)77.185.22.15 (talk) Starting with "For a given uniform (lux) light source, a given lens aperture and a given exposure time" leaves the initial conditions vague and inadequate. From the conclusion "Lumen and light energy is proportional to the pixel area and image sensor area" we can deduce that you are referring to a constant relative aperture, or f-number;(as F-number and aperture points out, people nearly always mean relative aperture. Please understand WHY! Important!)77.185.22.15 (talk) but then you don't say if you're letting the focal length (its independent > constant lux!)77.185.22.15 (talk) scale with the sensor size or not, so we don't know if the actual aperture, or total amount of light being captured from the scene, is varying with format or not. This matters a lot in some deductions; assuming constant focal length is not compatible with letting sensor size change,[citation needed] and assuming constant f-number is sort of a cheat, as it lets you use more light, at the cost of reduced DoF.(not specified!!!)77.185.22.15 (talk) One needs to at least be clear what assumptions one is comparing on. If you assume constant absolute aperture, as is appropriate in a system with DoF constraint,(not specified!!!)77.185.22.15 (talk) then the noise advantage will go to the smaller sensor, for a given number of pixels.
You claim that "Pixels should be independent and unrelated in an image sensor". That's a simplification of reality, since pixels in real images tend to fairly highly correlated.[citation needed] Nevertheless, we can live with it. More importantly, you need to assume that pixel noises are independent to do the kind of noise analysis that Bobn2 did; he had that right.[citation needed] And he included the crucial noise source read noise, a noise that doesn't depend on light level, that is needed to understand how noise and resolution trade when subdividing pixels; his analysis was over-simplified perhaps, in assuming a particular dependence of read noise on pixel size, but not bad.
I'm not following who said what or what fact you're proposing in the your last bullet point.
The problem, as I said before, with Bobn2's rewrite was it was too long and detailed and unsourced.
- I agree with that. It was a bit of a quick job, I saw the first paragraph, which I thought very misleading, particularly on the DR issue, and waded in. In fact, the two sources I gave, the DxO 'insights' and the Aptina white paper supported each point that I made, maybe I needed to be more assiduous at referencing. The reason it gets long is because there is so much disinformation on these matters which needs to be copuntered point by point, else people are wont just to revert to the misinformed materialBobn2 (talk) 14:26, 13 December 2011 (UTC)
Maybe we can move toward its correctness improvement in smaller steps, citing sources along the way, and end up in a better place. Dicklyon (talk) 16:00, 12 December 2011 (UTC)
- Agreed, let's start with the first paragraph, which is here:
- If all other performance aspects of a sensor's materials and fabrication techniques and amplification electronics are equal, a larger pixel size or pitch (i.e. lower resolution) sensor captures images with greater dynamic range than a smaller sensor.
- Note that what we are talking about here is 'capturing images' not pixels. And there is no source cited. So, a good source to cite is the DxO insight that I used, since that explains how DR scales with pixel count. Either that source is right, or it is wrong. If we wish to talk about 'images' with higher DR then we need to be making a comparison between images. That includes, as I said in my rewrite, comparing the images, not the pixels, or equivalently, comparing over the same bandwidth. The pixel size issue is irrelevant. Perhaps we could rewrite to:
- If all other performance aspects of a sensor's materials and fabrication techniques and amplification electronics are equal, a larger pixel size or pitch (i.e. lower resolution) sensor captures images with greater dynamic range than a smaller sensor.
- Agreed, let's start with the first paragraph, which is here:
- When capturing an image captured at some given exposure, the total amount of light (i.e. the number of photons) which make up the image is determined by the area of the sensor. If all other performance aspects of a sensor's materials and fabrication techniques and amplification electronics are equal then the signal to noise ratio will depend on the photon shot noise, which in turn is proportional to the square root of the number of photons. Thus, a larger sensor will tend to have have a higher signal to noise ratio for a fixed exposure, although it should be noted that the visible appearance of an image captured by different sized sensors with the same exposure will not be the same. If instead of the same exposure the constraint is the same visible appearance (the same depth of field and motion blur (shutter speed) then the advantage of larger sensors is cancelled out, since the exposure needed for the same appearance reduces as the square of the linear scale factor, precisely cancelling the extra area.
- The larger pixel's area provides a larger collection trap or "well", in which photon-freed electrons can accumulate, giving rise to a higher ratio ("dynamic range") between the accurately recorded maximum and accurately recorded minimum number of incident photons striking each pixel area. The ratio between the accurate maximum and accurate minimum number of incident photons, that a sensor pixel can accurately report as having struck it over a given exposure interval, establishes an outer boundary for the theoretical signal to noise ratio (i.e. dynamic range) of the sensor for a given output resolution.
- I would suggest we simply lose this, since the discussion really isn't relevant here. it might be if topic was pixel size, but it isn't, it is sensor size.
- I'll put those in, and then perhaps we can work through paragraph by paragraph.Bobn2 (talk) 14:26, 13 December 2011 (UTC)
- quick answers. For given DOF (I LIKE!) its another thing. 77.185.22.15 (talk) 16:15, 12 December 2011 (UTC) added something. hopefully you both think to understand my free lesson. 77.185.22.15 (talk) 03:05, 13 December 2011 (UTC)
- I find your writing impossible to follow, so the free lesson is going to waste. Dicklyon (talk) 05:52, 13 December 2011 (UTC)
- My approach now is to follow as you suggested, a stage by stage rewrite, so on to the next paragraph:
- It is true that pixel size or area places an limit on the maximum number of photons that can be recorded by a given pixel. It is also true that various kinds of inherent fluctuations in weak signals limit the minimum number of photons a smaller pixel may accurately record. But one cannot predict the maximum ratio between accurate maximum and accurate minimum photon counts of a sensor, one cannot predict the sensor's dynamic range, merely by noting the "largeness" of a sensor pixel. Because the nature and efficiency of a sensor pixel's materials, and its fabrication technology, and surrounding optically significant overlays and amplification and recording systems, also further restrict the accuracy of both the largest and smallest photon counts.
- Actually this is not true. The pixel size really is not closely correlated to the number of photons that can be collected by the pixel, to collect more photons the designer simply makes the capacitance higher, which can be achieved in several ways. If you look at the Aptina white paper, they have a switchable capacitance for just this reason. So, this paragraph is incorrect, it starts off saying something is 'true' which isn't. Again, it seems to me that the pixel size discussion is not directly related to the sensor size one, so shall we simply delete that paragraph? Bobn2 (talk) 14:45, 13 December 2011 (UTC)
- Bob, thanks for working on this. I think there's some truth to it (and the Aptina paper talks about adding capacitance only in the big-pixel case, where it's easy to do so). But I have no objection to simply removing things that are unsoured and disputed. There's a lot that's unsourced in the article, which Pol098 has now sensibly tagged as needing sources; feel free to prune to what you think is true and sourcable, but don't throw it all out at once. And don't try to articulate and counter all the misinformation that's out there; that's a losing game. Simply write what's true, and back it up with sources, and the article will be greatly improved. Dicklyon (talk) 16:06, 14 December 2011 (UTC)
- Spent a bit of time this time, and tried to find some good web accessible sources. I hope that is easier to read and more to the point. I'll move onto the diffraction part now.Bobn2 (talk) 18:43, 17 December 2011 (UTC)
- Bob, thanks for working on this. I think there's some truth to it (and the Aptina paper talks about adding capacitance only in the big-pixel case, where it's easy to do so). But I have no objection to simply removing things that are unsoured and disputed. There's a lot that's unsourced in the article, which Pol098 has now sensibly tagged as needing sources; feel free to prune to what you think is true and sourcable, but don't throw it all out at once. And don't try to articulate and counter all the misinformation that's out there; that's a losing game. Simply write what's true, and back it up with sources, and the article will be greatly improved. Dicklyon (talk) 16:06, 14 December 2011 (UTC)
Suggested re-ordering
Having redrafted the noise section, and looking on to the diffraction section, I see that both depend in some sense on the DOF discussion, which currently comes after them. Moreover. I suspect that DOF is the first difference many photographers think of when they discuss image sensor size, so perhaps this should be the first discussion. In any case, the DOF issue wrt different sensor formats is pretty much covered by the DOF article, so this section can be covered with a quick reference and precis. Then I would suggest a separate lens size discussion. Lens size and performance leads naturally into the diffraction discussion, then to the noise and DR. Seems a more natural order to me. — Preceding unsigned comment added by Bobn2 (talk • contribs) 19:52, 17 December 2011 (UTC)
- That sounds sensible. I haven't read all that you've done yet, but I made some tweaks to part of it and added a subsection to mention that there are different situations that people compare. Let me know if you agree it's OK. Dicklyon (talk) 23:46, 17 December 2011 (UTC)
- I'm happy with that. I thought about a bigger section on the bases for comparison, but it starts to get convoluted again. Your short section is a goof compromise. I decided to go with the different size/same pixel count since it extracts size as a independent variable. In practice for these comparisons, the pixel doesn't matter much, all that matters is taking like proportions of the image. Of course, if the sensor formats have different aspect ratios, like proportions becomes a little more tricky.Bobn2 (talk) 00:15, 18 December 2011 (UTC)
- Done the DOF section. The bit is between my teeth now, so I'll move on to lens size and diffraction.Bobn2 (talk) 17:40, 18 December 2011 (UTC)
- Lens size redrafted.Bobn2 (talk) 12:55, 19 December 2011 (UTC)
- Now I've done the diffraction section. I moved Eric Fossum's lecture to the links, since it did not directly address the diffraction issue.Bobn2 (talk) 23:28, 21 December 2011 (UTC)
Original research 2
I see a lot of unsubstantiated statements in this article. Some of them are quite reasonable, some are not. For example, let's concentrate on the following sentence: "The evolution of the classic four element 'Tessar' lens design is discussed by Nasse[10], and shows its evolution from an f/6.3 lens for plate cameras through to an f/2.8 5.2mm four element optic with eight extremely aspheric surfaces, economically manufacturable because of its small size. Its performance is 'better than the best 35 mm lenses – but only for a very small image'." It is apparent to anyone reading the cited article that:
- the evolution of the classic four element Tessar lens design is not discussed in Nasse, what is discussed in Nasse is the history of those lenses which have been marketed under the Tessar name,
- the statement about the performance of the tiny f/2.8 5.2mm lens is not backed by the article, what the article says is that the small lens has better resolution of any 35mm lens. While resolution is the technically correct term here, performance is not, and it suggestive of a false statement (that the small lens produces more detailed images of a given subject than any 35mm lens, which is not true).
The best correction I see for this case is the removal of the offending phrase, which I will carry out sometimes in the future, unless someone suggests a better solution. In general, I suspect that a non-negligible fraction of the article should be deleted. — Preceding unsigned comment added by 79.24.13.93 (talk) 22:53, 26 December 2011 (UTC)
- Good point, rather than delete it, I'll make the wording more precise to cover one of the points you make, that this is not the development of the original Tessar line. The statement that you attribute, 'that the small lens has better resolution of any 35mm lens' is not in the article, what is is the phrase 'better than the best 35 mm lenses – but only for a very small image' which is a verbatim quote from Nasse.Bobn2 (talk) 22:44, 27 December 2011 (UTC)
- Change made, I hope you approve of the amended wording.Bobn2 (talk) 22:49, 27 December 2011 (UTC)
- Well, both phrases are in the article, actually. That about resolution is in the text, page 6. That about performance is in a caption, page 9. Nevertheless, by being too picky, I ended up obscuring my opinion... We must keep in mind that the phrase we are discussing is meant to clarify the previous one, which is: "to maintain the same absolute amount of information in an image the lens for a smaller sensor requires a greater resolving power". To this aim, the current wording is, I think, too verbose, so much that its very purpose may not be apparent. I suggest to replace the whole sentence with the following "For example, a lens designed for camera phones may have much better resolution than even the best lenses for 35mm photography [cite Nasse]". What do you think about that? — Preceding unsigned comment added by 79.24.13.93 (talk) 00:53, 28 December 2011 (UTC)
- I think that if you want to make such a comparison, you need to go to a little trouble to make absolutely sure the meaning in clear, by saying in the image space, as opposed to the object space, or something like that. Shrinking a lens design does shrink its aberrations, but preserves its Airy disk, so the image space resolution can get better for a while, but will get worse again with too much scaling. Dicklyon (talk) 05:11, 28 December 2011 (UTC)
- The Airy disc issue was covered under diffraction, I think. Bobn2 (talk) 16:01, 28 December 2011 (UTC)
- Yes, but it's the key to this scaling issue, I think. Scaling down at constant f-number makes the object resolution worse when you're in a diffraction-limited regime; but it stays the same in an aberration-limited regime, if you don't get much closer to diffraction limited. And the text is correct, but will be hard for many to correctly understand, I think, where it says "to maintain the same absolute amount of information in an image (which can be measured as the space bandwidth product[9]) the lens for a smaller sensor requires a greater resolving power" and "better than the best 35 mm lenses – but only for a very small image." The notions of "resolving power" and "better" are focal-plane or image-space concepts here, which is obvious to you and to me, but perhaps not to general readers. The implications of this are that smaller lenses usually need to be designed with lower f-numbers. Dicklyon (talk) 16:07, 28 December 2011 (UTC)
- I would tend to reverse the order. 'Smaller lenses need to be designed with lower f-numbers' is a corollary of the 'same photo' constraint, essentially that diffraction burring for any given DOF is sensor size invariant, thus to achieve the same diffraction limited resolutions it is necessary to have the same DOF, and hence lower f-numbers. Then the scaling paradigm goes with the 'same exposure' constraint, and inherently means less resolution for smaller sensors. The 'same lens' constraint makes no sense in this section. Then we come to the issue of how smaller f-numbers might be provided, and that can discuss the availability of different manufacturing techniques as size is reduced - as illustrated by Nasse. I'm loath to lose that reference. It is a very clear and example rich source. Bobn2 (talk) 13:38, 29 December 2011 (UTC)
- Yes, but it's the key to this scaling issue, I think. Scaling down at constant f-number makes the object resolution worse when you're in a diffraction-limited regime; but it stays the same in an aberration-limited regime, if you don't get much closer to diffraction limited. And the text is correct, but will be hard for many to correctly understand, I think, where it says "to maintain the same absolute amount of information in an image (which can be measured as the space bandwidth product[9]) the lens for a smaller sensor requires a greater resolving power" and "better than the best 35 mm lenses – but only for a very small image." The notions of "resolving power" and "better" are focal-plane or image-space concepts here, which is obvious to you and to me, but perhaps not to general readers. The implications of this are that smaller lenses usually need to be designed with lower f-numbers. Dicklyon (talk) 16:07, 28 December 2011 (UTC)
- The Airy disc issue was covered under diffraction, I think. Bobn2 (talk) 16:01, 28 December 2011 (UTC)
- My thinking in including the Nasse reference, and the statement about the development of the Tessar lens was to underline the point that not only was higher resolution required from small format lenses, but also that their development would not simply be a process of scaling, but adoption of different optical fomulae, a point rather nicely made by the Nasse article, I thought. To be honest, I'm at a bit of a loss to know what is the nub of your objection. My quote from Nasse is pretty much in context and convey's his meaning. The full caption is 'Performance data of a Tessar lens for a mobile phone camera measures with 20, 40 and 80 Lp/mm. It is better than the best 35 mm lenses – but only for a very small image.' I find it difficult to relate the 'it' to anything other than 'performance', if it is the word 'performance' that you take issue with. Now, if you are saying that the overall performance of a camphone with this lens is unlikely to exceed the performance of a larger sensor camera, then you are right. But I don't see how one would infer than meaning from what is written. As it is, I am loath to use your suggested wording, which removes the context of the design process not being one of simple scaling. Anyway, while we are at it, your original comment cited this only as an instance of 'a lot of unsubstantiated statements in this article'. If you could be explicit about which statements you find unsubstantiated, we could go about substantiating them, or removing them if they are not substantiateable.Bobn2 (talk) 15:58, 28 December 2011 (UTC)
- I think that if you want to make such a comparison, you need to go to a little trouble to make absolutely sure the meaning in clear, by saying in the image space, as opposed to the object space, or something like that. Shrinking a lens design does shrink its aberrations, but preserves its Airy disk, so the image space resolution can get better for a while, but will get worse again with too much scaling. Dicklyon (talk) 05:11, 28 December 2011 (UTC)
- As for unsubstantiated stuff, its all over the place. There are whole sections like "Sensor size and telecentricity" or "Bridging the gap in sensor sizes" which do not cite a single reference. As for the Tessar stuff, I think that Dicklyon described the point very clearly. The correct term is "image space resolution". The quote form Nasse is misused, because you are quoting an informal statement out of its context. In fact, the statement is the caption of a graph, and it is meant to make sense only to those people that actually see the graph (and know how to read it), that is its context and it can not be taken out of it. In the text of Nasse's article itself (not in a caption), you will find a more precise statement, and there he says "resolution". Also, this paper (Nasse) does not appear to be a peer-reviewed scientific publication, it looks more as some sort of Zeiss divulgation thing (no bibliography, very informal, the name of the author himself is not there, just "H. H. Nasse"). But again, all this is unnecessarily verbose and baroque. The entire section basically says "It might be expected that a lens for a big sensor could be scaled down to fit a smaller sensor, however this is not true for a million reasons" (and it doesn't even mention diffraction). Ok, so what? It might be expected that a truck could be scaled down to make a car, however this is not true for a million reasons: why don't we make a section like this in the Wikipedia article about cars? Because encyclopedic articles are about facts, not about non-facts. The last few sentences, after "In summary", look like factual to me, but I would kill all the rest. — Preceding unsigned comment added by 79.11.75.239 (talk) 05:33, 29 December 2011 (UTC)
- The unsubstantiated stuff is what I'm engaged in reworking, which mostly involves rewriting because it's unsubstantiated and wrong (therefore unsubstantiable). The Telecentricity section was what I was going to do next, since 'telecentricity' is not at all connected with sensor size, I was going to rework it to look at the differential effect of f-number coupling between microlens and objective lens (which is pixel size, and therefore sensor size, on an assumption of equal pixel counts, to a degree). I don't have a lot of sympathy with your point of view on the lens size stuff, because if this section is going to be other than an account of some very abstruse scaling theory, it will survey industry practice in the provision of lenses for different formats, which is something that won't be found in the formal research literature. Nasse is a good source for that, because it least it is attributable and well written, although of course we have to be aware of its provenance. If you know of a better survey of industry practice, I'd be glad to use it. Still, given that you have picked this section out for critique, it's clear it needs some more work. I'll try another redraft and see if that suits better. Bobn2 (talk) 13:26, 29 December 2011 (UTC)
- As for unsubstantiated stuff, its all over the place. There are whole sections like "Sensor size and telecentricity" or "Bridging the gap in sensor sizes" which do not cite a single reference. As for the Tessar stuff, I think that Dicklyon described the point very clearly. The correct term is "image space resolution". The quote form Nasse is misused, because you are quoting an informal statement out of its context. In fact, the statement is the caption of a graph, and it is meant to make sense only to those people that actually see the graph (and know how to read it), that is its context and it can not be taken out of it. In the text of Nasse's article itself (not in a caption), you will find a more precise statement, and there he says "resolution". Also, this paper (Nasse) does not appear to be a peer-reviewed scientific publication, it looks more as some sort of Zeiss divulgation thing (no bibliography, very informal, the name of the author himself is not there, just "H. H. Nasse"). But again, all this is unnecessarily verbose and baroque. The entire section basically says "It might be expected that a lens for a big sensor could be scaled down to fit a smaller sensor, however this is not true for a million reasons" (and it doesn't even mention diffraction). Ok, so what? It might be expected that a truck could be scaled down to make a car, however this is not true for a million reasons: why don't we make a section like this in the Wikipedia article about cars? Because encyclopedic articles are about facts, not about non-facts. The last few sentences, after "In summary", look like factual to me, but I would kill all the rest. — Preceding unsigned comment added by 79.11.75.239 (talk) 05:33, 29 December 2011 (UTC)
- Now I've had a go at 'sensor size and telecentricity'. I can't see the argument that telecentricity is impacted by sensor size at all (except insofar as it makes lenses big, so there's more scope for oversize lenses than with a small sensor). Instead I've focussed on the f-number shading effect, which does seem demonstrably to get worse as pixel size reduces (and is therefore linked indirectly to sensor size) Bobn2 (talk) 17:56, 29 January 2012 (UTC)
Aspect Ratio
Aspect ratio is not mentioned in the artcle.
There should be a special section for aspect ratio.--98.199.22.63 (talk) 05:47, 2 February 2012 (UTC)