How about Dynamic range?

[mailman88]

...Thats a fact and true---> The camera does not recover any highlight details in the raw data.

 

[Steen Bay]

...Thats a fact and true---> The camera does not recover any highlight details in the raw data.

But the RAW data has more highlight (and shadow) information than the in-camera JPEGs, and that's of course what people are refering to when they are talking about "headroom" and "recovery". That's why every DPR (DSLR) review has a section about "RAW headroom", telling us that "..there is typically around 1 EV of extra information available at the highlight end in RAW files..", which enables us to ".. recover detail lost to over-exposure."

 

[Steen Bay]

Past comments and current impressions are made from past cameras. Based on the numbers I just showed, however, it seems Canon may be clipping more headroom in the 125/250 group than previously (they used the full values up to 16383 before subtracting blackpoint), and now the 160/320 group have more DR (but less headroom) than the "main" ISOs. I gotta go now, I'll look into it more when I get home later.

I just did the math, and it seems 200 and 160 have exactly the same DR, so no change there.

Same DR as ISO 200, but 1/3 stop less 'headroom' (and 1/3 stop more 'footroom') at ISO 160, right? (And same headroom at ISO 200 and 250?) And btw, I noticed that Andy had a remark in the 100/2.8L IS Macro review about "..the 50D's relatively aggressive optical low-pass filter." (page 3), but isn't your impression the opposite, that the 50D has a relatively weak AA-filter?

 

[John Sheehy]

Past comments and current impressions are made from past cameras. Based on the numbers I just showed, however, it seems Canon may be clipping more headroom in the 125/250 group than previously (they used the full values up to 16383 before subtracting blackpoint), and now the 160/320 group have more DR (but less headroom) than the "main" ISOs. I gotta go now, I'll look into it more when I get home later.

I just did the math, and it seems 200 and 160 have exactly the same DR, so no change there.

Same DR as ISO 200, but 1/3 stop less 'headroom' (and 1/3 stop more 'footroom') at ISO 160, right?

Yes.

(And same headroom at ISO 200 and 250?)

Apparently, but now that we have auto-ISO in manual mode, there is another way we must look at it. If the exposure is manual, and the ISO is floating, then the 125 group will be the only ones that can be bad, because with a fixed manual exposure, it doesn't matter whether the camera chooses 160, 200, or 250, noise-wise, because there is no difference. The only difference is that the 250 will be clipped. Unfortunately, the camera has no feature to use full stops of ISO in manual exposure. The CF setting is ignored in this mode.

And btw, I noticed that Andy had a remark in the 100/2.8L IS Macro review about "..the 50D's relatively aggressive optical low-pass filter." (page 3), but isn't your impression the opposite, that the 50D has a relatively weak AA-filter?

Well, I get better pixel-level sharpness with my 50D than my 30D, with a sharp lens. Here's a 100% crop of my 50D with a 90mm Tamron and a 1.4x TC:



I sharpened this a bit on the heavy side, perhaps, but you can see that the image is aliased (IOW, the under-sampling artifacts sharpen long before noise does).

--
John

 

[GaborSch]

According to DXOmark's DR, the D3X is pretty linear above 400, and 200 has about 1/4 stop more read noise, and 100 about 1/2.

I don't have any D3X images suitable for such measurement, but I doubt anyway, that the D3X has any relevance regarding the A900. The sensor may be the same, but the electronics is not. This is not all, but it is enough to dismiss any such claim regarding the A900 based on the D3X.

When measured on A900 black frames, there is nothing even close to the assessments posted by pixelpeeper.

All this is not important, I was only wondering how those numbers have been arrived at.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

 

[BMA]

The pros outwiegh the cons with this form of compession but going from
14 (or more) to 8 bits would have to compromise quality to some degree
giving you less grading latitude.

All other things being equal. Adding more bits to an analog to digital conversion process only changes the exactness of the captured data. This does have it's benefits. When you apply a curve translation to the data there is more information to use for more precisely generating the output. This makes it easier to take a part of the input range and expand it's DR in the output. Sure you do loose more information when going from 14 bits to 8 bits than from 12 bits to 8 bits, but you gain in the exactness of hue and saturation able to be transfered from the input to output and that is what counts. That extra exactness of hue and saturation could be used to add more contrast to the clouds and bring out detail in the shadow regions.

--
Bryan - click, click, click, click, moo, click, click...

 

[pixelpepper]

According to DXOmark's DR, the D3X is pretty linear above 400, and 200 has about 1/4 stop more read noise, and 100 about 1/2.

I don't have any D3X images suitable for such measurement,

Most of the measurements on the Sony column ADC sensors (Clark, Martinec, Sheehy, DxO as interpreted by DSPograher) seem to triangulate quite will, with read noises of 6-9 e- at base ISO and 4-5 e- at highest 'real' ISO. Yours might be different, if so you're an outlier, which doesn't necessarily mean you're wrong.

but I doubt anyway, that the D3X has any relevance regarding the A900. The sensor may be the same, but the electronics is not.

Given that the sensor has a digital output, there is no scope for the 'electronics' to be functionally different. What is most likely different is the various signal processing algorithms that Nikon uses and Sony doesn't. The architecture would seem to allow quite a lot of scope for signal processing improvements, either using Nikon special sequencing microcode on the sensor or not,

1. Use of multiple sampling and CDS in the signal processor, rather than Sony's 'count down-count up' method of CDS.
2. Use of multiple sampling and sample averaging to reduce noise.
3. Identification and correction of systematic data errors.
4. Slower sampling to reduce noise bandwidth.
5. Changing sampling sequence to extend settling time.

And probably a few more that Nikon engineers, with familiarity with the circuitry, could come up with.

The most probable 'electronic' improvement is better power supply regulation to improve reset noise.

This is not all, but it is enough to dismiss any such claim regarding the A900 based on the D3X.

Not at all, since the sensor is essentially the same, the D900 forms the baseine for the D3x.

When measured on A900 black frames, there is nothing even close to the assessments posted by pixelpeeper.

I would be interested in knowing your figures. Emil's measurements on the D300 (which has the same A to D arrangements) put it close at base ISO. I did rather over egg the pudding, saying Canon style cameras had 10x the read noise at base ISO, more like 3-5x, if the various figures are right.

 

[John Sheehy]

The pros outwiegh the cons with this form of compession but going from
14 (or more) to 8 bits would have to compromise quality to some degree
giving you less grading latitude.

All other things being equal. Adding more bits to an analog to digital conversion process only changes the exactness of the captured data. This does have it's benefits. When you apply a curve translation to the data there is more information to use for more precisely generating the output. This makes it easier to take a part of the input range and expand it's DR in the output. Sure you do loose more information when going from 14 bits to 8 bits than from 12 bits to 8 bits, but you gain in the exactness of hue and saturation able to be transfered from the input to output and that is what counts. That extra exactness of hue and saturation could be used to add more contrast to the clouds and bring out detail in the shadow regions.

What you say sounds good in theory, but would really only be applicable if we were talking 8 bits RAW vs 10 bits; the difference between 12, 14, and 16 bits is irrelevant unless manufacturers can stop adding so much noise to the signal, in the ADC and other late-stage parts of the signal chain.

The fact is, most of those extra two bits are just noise. A careful converter does not need them. A sloppy converter may benefit simply because it may force more working precision, possibly resulting in less posterized output. A typical Canon 14-bit ISO 100 RAW has so much noise that the value of black varies over 17 RAW levels, and the photon noise of highlights makes the values vary even greater than that (hundreds of RAW levels), for a so-called smooth, OOF subject.

No Canon's RAW data would be posterized in a 12-bit RAW, and you only need 12 bits for low ISOs and their deepest shadows. For a high-key ISO 6400 image, you could literally get away with just 4 or 5 bits. Remember the late 80s when 256-color GIFs were emulating 24-bit graphics with floyd-steinberg diffusion dithering? Same principle, except you're not adding the noise; it is always there, anyway, and unavoidable.

--
John

 

[John Sheehy]

All this is not important, I was only wondering how those numbers have been arrived at.

It would be nice if someone with the resources could host a site where RAW blackframes, OOF color-checkers, clipped files, and other such useful files could be hosted.

--
John

 

[GaborSch]

Most of the measurements on the Sony column ADC sensors (Clark, Martinec, Sheehy, DxO as interpreted by DSPograher) seem to triangulate quite will, with read noises of 6-9 e- at base ISO and 4-5 e- at highest 'real' ISO. Yours might be different, if so you're an outlier, which doesn't necessarily mean you're wrong

I don't have any electron counter. If I say measuring , then I mean using the raw data (digital), which is the product of the camera.

In case of Canon (and a few other) cameras I measure the noise on the masked pixels. Neither the D3X nor the A900 pass this info (or there is none on the sensor?), so I can use only black frames. The Canon sensors show, that the black frame measurements are very close to the masked areas.

Given that the sensor has a digital output, there is no scope for the 'electronics' to be functionally different

1. the D3X has lower noise,

2. the D3X can create about 14000 levels (in 14bit mode, roughly), the A900 creates about 3000 levels.

I don't discuss in hardware issues, for I don't know of it at all, but are you saying, that this is the same electronics?

However, all that is on the sideline, the topic is read noise.

I would be interested in knowing your figures. Emil's measurements on the D300 (which has the same A to D arrangements) put it close at base ISO. I did rather over egg the pudding, saying Canon style cameras had 10x the read noise at base ISO, more like 3-5x, if the various figures are right.

Again, I can talk only about the noise measured on the raw data. Let's see same examples, expressed as the standard deviation in percentage of the numerical pixel value range (clipping level minus black level). The standard deviation in absolute values is not useful, because the pixel value range changes with ISO, or in case of Nikons, with bit depth.

Canon 40D:

ISO 100: 0.039%, ISO 200: 0.044%, ISO 400: 0.060%, ISO 800: 0.1%, ISO 1600: 0.16%

Canon 5D2:

ISO 100: 0.038%, ISO 200: 0.038%, ISO 400: 0.042%, ISO 800: 0.050%, ISO 1600: 0.068%

Nikon D300:

ISO 100: 0.027%, ISO 200: 0.033%, ISO 400: 0.062%, ISO 800: 0.122%, ISO 1600: 0.24%

All the above was measured on the masked areas. The Sony A900 values are from black frames, the green channel:

ISO 100: 0.031%, ISO 200: 0.034%, ISO 400: 0.066%, ISO 800: 0.126%, ISO 1600: 0.24%

NOTES:

1. The measured values can differ between camera copies, or even between shots of the same copy, by more than 10% (sometimes 20%)

2. I measured the NOT BL corrected values, as only those represent the true noise. This is not possible with Nikons.

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

 

[pixelpepper]

Most of the measurements on the Sony column ADC sensors (Clark, Martinec, Sheehy, DxO as interpreted by DSPograher) seem to triangulate quite will, with read noises of 6-9 e- at base ISO and 4-5 e- at highest 'real' ISO. Yours might be different, if so you're an outlier, which doesn't necessarily mean you're wrong

I don't have any electron counter. If I say measuring , then I mean using the raw data (digital), which is the product of the camera.

The sensor is an electron counter. From the raw readings you can derive the values you need. Emil discusses the method on his pages. Essentially, by measuring the shot noise at a given illumination, you can calculate the number of electrons per pixel that that level of illumination represents, since the standard deviation of the shot noise is the square root of the number of electrons. Scaling that to full count gives the saturation count, since sensors a pretty linear. Saturation divided by full count gives the 'gain', The black level count gives the read noise, which can be scaled to electrons using the 'gain'. Nikons are a bit more tricky, since they clip the blacks.

In case of Canon (and a few other) cameras I measure the noise on the masked pixels. Neither the D3X nor the A900 pass this info (or there is none on the sensor?), so I can use only black frames. The Canon sensors show, that the black frame measurements are very close to the masked areas.
Should be, if the black frame is black.

Given that the sensor has a digital output, there is no scope for the 'electronics' to be functionally different

1. the D3X has lower noise,

2. the D3X can create about 14000 levels (in 14bit mode, roughly), the A900 creates about 3000 levels.

I don't discuss in hardware issues, for I don't know of it at all, but are you saying, that this is the same electronics?

Yes, substantively. The chip give a digital output. There is no scope for different electronics, unless the chip is very different (and there would be no point in it being so similar if it were very different). The signal processing functions may well be very different, but signal processing can equally be applied in the analog domain, using 'different electronics' or in the digital domain, using different digital functions. In electronic design, there has been a steady drift from the analog domain to the digital domain, which has the advantage of being modifiable in firmware. One good indicator that this is happening in the digital domain is that the D3x takes roughly three times the time to create its 14000 levels as does the A900 to create its 3000 levels. I also indicated several ways in which 'different noise' could be produced in the digital domain.

However, all that is on the sideline, the topic is read noise.

I would be interested in knowing your figures. Emil's measurements on the D300 (which has the same A to D arrangements) put it close at base ISO. I did rather over egg the pudding, saying Canon style cameras had 10x the read noise at base ISO, more like 3-5x, if the various figures are right.

Again, I can talk only about the noise measured on the raw data. Let's see same examples, expressed as the standard deviation in percentage of the numerical pixel value range (clipping level minus black level). The standard deviation in absolute values is not useful, because the pixel value range changes with ISO, or in case of Nikons, with bit depth.

Canon 40D:

ISO 100: 0.039%, ISO 200: 0.044%, ISO 400: 0.060%, ISO 800: 0.1%, ISO 1600: 0.16%

Canon 5D2:

ISO 100: 0.038%, ISO 200: 0.038%, ISO 400: 0.042%, ISO 800: 0.050%, ISO 1600: 0.068%

Nikon D300:

ISO 100: 0.027%, ISO 200: 0.033%, ISO 400: 0.062%, ISO 800: 0.122%, ISO 1600: 0.24%

All the above was measured on the masked areas. The Sony A900 values are from black frames, the green channel:

ISO 100: 0.031%, ISO 200: 0.034%, ISO 400: 0.066%, ISO 800: 0.126%, ISO 1600: 0.24%

Interesting. They do indeed indicate the flat read noise characteristic of the Sony sensors (since the full count reduces by half with each doubling in ISO) They also show the falling read noise characteristic of the Canon sensors. Scaled to the different full capacity of the pixels, they also triangulate well with the other measurements, so I think yours seem quite consistent to me.

http://www.panopeeper.com/panorama/pano.htm

 

[GaborSch]

The sensor is an electron counter. From the raw readings you can derive the values you need. Emil discusses the method on his pages

I know the method, but I don't accept it. It works only under the assumption, that the read noise is constant (with a given ISO), but it is not.

since the standard deviation of the shot noise is the square root of the number of electrons

The standard deviation is the noise, and the shot noise part of it changes proportionally to the square root of the change of illumination.

Anyway.

They do indeed indicate the flat read noise characteristic of the Sony sensors

Well, yes, from ISO 400 to 1600, but not from ISO 100 to 400.

Scaled to the different full capacity of the pixels, they also triangulate well with the other measurements, so I think yours seem quite consistent to me.

1. These values can not be and don't need to be scaled. They are the percentage of the noise in the numerical range .

2. You wrote in the earlier message The Sony column ADC system seems to have about 50% more read noise at low ISO than at high ISO (which compares with 10x or more in ost conventional systems) , and that is, what I had problem with.

The A900's noise with ISO 1600 is 0.24%, with ISO 100 it is 0.031. 1/16th of the ISO 1600 noise is 0.015%, i.e. the ISO 100 noise is 100% greater, not 50%.

However, that was not my problem. Rather, I don't understand your point. It sounds, like it would be a disadvantage, that the noise at low iso is more, than it is "supposed to be" based on the noise with higher ISO. But why? I see it the other way: the noise with ISO 1600 is "supposed to be" 16 times higher than with ISO 100, but it is only four times higher (40D), only 1.8 times higher (5D2), nine times higher (D300) and 7.7 times higher (A900).

Now, we have lots of nice number, but what for? I'm afraid they are good only for discharging the battery of my pocket calculator. LOL

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

 

[zappa1976]

All this is not important, I was only wondering how those numbers have been arrived at.

It would be nice if someone with the resources could host a site where RAW blackframes, OOF color-checkers, clipped files, and other such useful files could be hosted.

--
John

Here:

http://www.mediafire.com/?sharekey=f7be5d1881c0faddab1eab3e9fa335ca59273cf1b5086b04

I'm uploading (pls wait some time because my adsl is very slow today!) 2 blackframes (100 and 200 iso) of a D3X and a blackframe at 100 iso of my 1Ds3.
I hope these could be useful for some analysis.
Pls can you tell me/us what do you think about these files?
Thanks.

P.S.: here are available, at the moment, from the D3X:
http://www.zshare.net/download/684056477a40b936/
http://www.zshare.net/download/684064164b054e88/

 

[pixelpepper]

The sensor is an electron counter. From the raw readings you can derive the values you need. Emil discusses the method on his pages

I know the method, but I don't accept it. It works only under the assumption, that the read noise is constant (with a given ISO), but it is not.

Wrong. It works under the assumption that read noise is negligible. The point is to do the white frame test at an illumination much higher than the dark level. If we look at your measurements for the A900, at ISO 100 the read noise is 0.031% of full scale. Full scale is, of course 100%. If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced.

since the standard deviation of the shot noise is the square root of the number of electrons

The standard deviation is the noise

erm, no, the standard deviation is the standard deviation of the noise,

and the shot noise part of it changes proportionally to the square root of the change of illumination.

So long as you keep your units of illumination consistent, the standard deviation of the shot noise is the square root of the level of illumination. There is no constant of proportionality involved, whether your units are electrons, ADU or percentage of full scale.

Anyway.

They do indeed indicate the flat read noise characteristic of the Sony sensors

Well, yes, from ISO 400 to 1600, but not from ISO 100 to 400.

100 is the problem, since 100 is not a 'real ISO', it is just an overexposure of the base ISO. 200 onwards fit on a line.

Scaled to the different full capacity of the pixels, they also triangulate well with the other measurements, so I think yours seem quite consistent to me.

1. These values can not be and don't need to be scaled.

Anything can be scaled, all scaling does is change the units. Scaling to full capacity places the read noise in 'electrons', since the number of electrons represented by full scale changes with the ISO.

They are the percentage of the noise in the numerical range .

I understand that.

2. You wrote in the earlier message The Sony column ADC system seems to have about 50% more read noise at low ISO than at high ISO (which compares with 10x or more in most conventional systems) , and that is, what I had problem with.

Yes, I acknowledged that the 10x was an over statement when it comes to the Canon sensors.

The A900's noise with ISO 1600 is 0.24%, with ISO 100 it is 0.031. 1/16th of the ISO 1600 noise is 0.015%, i.e. the ISO 100 noise is 100% greater, not 50%.

You are dealing with different full scales. It is 0.24% of a different total to the 0.031%, therefore the noise percentages are not comparable, but you have scaled (as you said was impossible/unnecessary), but, as I said, 100 is not a 'real ISO', so is not 1600 multiplied by 16.

However, that was not my problem. Rather, I don't understand your point. It sounds, like it would be a disadvantage, that the noise at low iso is more, than it is "supposed to be" based on the noise with higher ISO. But why? I see it the other way: the noise with ISO 1600 is "supposed to be" 16 times higher than with ISO 100, but it is only four times higher (40D), only 1.8 times higher (5D2), nine times higher (D300) and 7.7 times higher (A900).

Fine at high ISO, but at low ISO's there is more read noise than there needs to be. If the read noise was the same as it is at high ISO, there would be more DR at high ISO's, nor would there be any need for variable gain amplification, all ISO adjustment could be done by digital scaling.

Now, we have lots of nice number, but what for? I'm afraid they are good only for discharging the battery of my pocket calculator. LOL

If you want to know what the visible effect of noise is likely to be, it's sensible to reference it to the same units as the image, i.e. photoelectrons (converted photons). ROTFLMAO.

 

[ejmartin]

The sensor is an electron counter. From the raw readings you can derive the values you need. Emil discusses the method on his pages

I know the method, but I don't accept it. It works only under the assumption, that the read noise is constant (with a given ISO), but it is not.

Wrong. It works under the assumption that read noise is negligible. The point is to do the white frame test at an illumination much higher than the dark level. If we look at your measurements for the A900, at ISO 100 the read noise is 0.031% of full scale. Full scale is, of course 100%. If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced.

Right in principle, just a little correction on details:

Suppose the pixel holds 40000 electrons at RAW saturation. 50% is 20000 electrons, the read noise is 141 electrons, which is not 7% of 40000 (more like 3.5%). In Poisson statistics the fluctuation is only the square root of the average in the appropriate units, in this case electrons. Read noise .031% of saturation is 12.4 electrons. Total noise at 50% saturation is thus sqrt[141^2+12.4^2]=141.5

Still a negligible correction at the level of accuracy of typical measurements, but just wanted to set the record straight.

So long as you keep your units of illumination consistent, the standard deviation of the shot noise is the square root of the level of illumination. There is no constant of proportionality involved, whether your units are electrons, ADU or percentage of full scale.

Not so, for the reason given above. There is a constant of proportionality involved if working in units other than electrons. That is how the gain enters -- it accounts for why the shot noise is not just the sqrt of the RAW level in ADU.

--
emil
--



http://theory.uchicago.edu/~ejm/pix/20d/

 

[pixelpepper]

The sensor is an electron counter. From the raw readings you can derive the values you need. Emil discusses the method on his pages

I know the method, but I don't accept it. It works only under the assumption, that the read noise is constant (with a given ISO), but it is not.

Wrong. It works under the assumption that read noise is negligible. The point is to do the white frame test at an illumination much higher than the dark level. If we look at your measurements for the A900, at ISO 100 the read noise is 0.031% of full scale. Full scale is, of course 100%. If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced.

Right in principle, just a little correction on details:

Suppose the pixel holds 40000 electrons at RAW saturation. 50% is 20000 electrons, the read noise is 141 electrons, which is not 7% of 40000 (more like 3.5%). In Poisson statistics the fluctuation is only the square root of the average in the appropriate units, in this case electrons. Read noise .031% of saturation is 12.4 electrons. Total noise at 50% saturation is thus sqrt[141^2+12.4^2]=141.5

Still a negligible correction at the level of accuracy of typical measurements, but just wanted to set the record straight.

So long as you keep your units of illumination consistent, the standard deviation of the shot noise is the square root of the level of illumination. There is no constant of proportionality involved, whether your units are electrons, ADU or percentage of full scale.

Not so, for the reason given above. There is a constant of proportionality involved if working in units other than electrons. That is how the gain enters -- it accounts for why the shot noise is not just the sqrt of the RAW level in ADU.

Thanks, Emil. You are, of course, right - we are dealing with the statistics of the electrons, not the derived units.

 

[GaborSch]

The standard deviation is the noise

erm, no, the standard deviation is the standard deviation of the noise,

The standard deviation is the standard deviation of whatever you are measuring. As we are talking about digital photography, we can conduct measurements only on the digital image, which consists of pixel values . The noise can be expressed as the standard deviation of those pixel values measured on a totally uniform area, i.e. on which the pixel values would be equal if there were no noise.

Wrong. It works under the assumption that read noise is negligible

Yes, it would work, but the read noise is not negligabe in my books. So, it works if either the full sensel capacity is known (I don't know that), or if the read noise is constant. However, the read noise is not constant based on my measurements + calculations.

If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced

(taking 3.5% instead of 7%)

ISO 100 with the A900, 50% intensity, the total measured noise is 0.5% (the read noise at this level is negligable). So much to the usefulness of this calculation.

100 is the problem, since 100 is not a 'real ISO', it is just an overexposure of the base ISO. 200 onwards fit on a line

100 is an overexposure of 160. Still it is out of line (the read noise with ISO 200 should be 1.26x that with ISO 160).

If you want to know what the visible effect of noise is likely to be, it's sensible to reference it to the same units as the image, i.e. photoelectrons (converted photons). ROTFLMAO.

Hurrrah. I am waiting for the next review posted on this site, stating the noise of the 3D with ISO 6400 is 273.6 photoelectrons .

--
Gabor

http://www.panopeeper.com/panorama/pano.htm

 

[ejmartin]

Yes, it would work, but the read noise is not negligabe in my books. So, it works if either the full sensel capacity is known (I don't know that), or if the read noise is constant. However, the read noise is not constant based on my measurements + calculations.

How do you think it varies? You say the read noise is not constant; what observations lead you to that conclusion?

If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced

(taking 3.5% instead of 7%)

ISO 100 with the A900, 50% intensity, the total measured noise is 0.5% (the read noise at this level is negligable). So much to the usefulness of this calculation.

I corrected pxlppr's math. 20000 electrons at 50% intensity would lead to 141 electrons of photon noise, or .7%, not 7%. Conversely, .5% noise at 50% intensity means that S/sqrt=200, or S=40000 electrons.

--
emil
--



http://theory.uchicago.edu/~ejm/pix/20d/

 

[pixelpepper]

If we do our light frame test at around 50%, then the shot noise will be 7.071%. If we read both the shot noise and read noise, the combined noise is sqrt(50+0.031^2) = 7.071%. The error in the reading is insignificant, so your mistrust of the method is misplaced

(taking 3.5% instead of 7%)

ISO 100 with the A900, 50% intensity, the total measured noise is 0.5% (the read noise at this level is negligable). So much to the usefulness of this calculation.

Emil corrected my mistake - you do need to do the sums in electrons, see his note.

100 is the problem, since 100 is not a 'real ISO', it is just an overexposure of the base ISO. 200 onwards fit on a line

100 is an overexposure of 160. Still it is out of line (the read noise with ISO 200 should be 1.26x that with ISO 160).

Sure, the curve is not absolutely straight. No-one said it was.

If you want to know what the visible effect of noise is likely to be, it's sensible to reference it to the same units as the image, i.e. photoelectrons (converted photons). ROTFLMAO.

Hurrrah. I am waiting for the next review posted on this site, stating the noise of the 3D with ISO 6400 is 273.6 photoelectrons .

It might be more difficult for the uninformed to interpret, but it would provide a consistent metric which could be used reliably to compare between cameras, unlike what they do now.

 

[Nello]

LOL dude 2 years of experience in Digital world is A LOT trust me! I've done tens of shoots...what do you do instead??? oh yeah I know...blogspot w awful pics..some underexposed...some waaay to bright and you're giving me lectures?LOL pleeeeeease take a walk buddy!
Btw nice hummer shot...almost burned my eyes LOL
http://xclusix.blogspot.com/
Here's the way to do it...i'm available for lessons on exposure LOL

I mean, he has have SO much experience

http://forums.dpreview.com/forums/read.asp?forum=1025&message=25379945

almost 2 years of experience since his first studio shot, so he knows about this magical apparatus called "camera"

(FOR THE OP: the 50D has about 2/3 stop more headroom in RAW, but a clear disadvantage on JPEG, about 2 stops, but thats not a sensor issue, its a software problem, easily corrected when converting RAW).
--
http://xclusix.blogspot.com