Dynamic range of some model cameras

Expose properly, negate the problem

On a serious note, though, would a test for tonal range breadth be like the test thing they do on dpreview, with the gradiated bar? I guess the best way i can think of would be to have the bar going from 0%-100% grey in segments, with each segment having a black bar above, and a white bar below. Check in PS where the furthest points apart lie that you can still see a difference between the grey and the black/white, and you have your dynamic range, no?

Or am i thinking something completely different?

davesrose wrote in post #3661554
well apart from the obvious in the table that Tom showed (do you see that you only have 15 shades of tone past stop 4?),

Tom's table showed that 8 bits have a MAXIMUM dynamic range of 8 stops, and a more feasible dynamic range (while still retaining some brightness resolution within each stop) less than that.

You stated that jpg was limited to 4 stops, or "little more than 4 stops." You clearly are not a mathematician, but "less than 8" is not the same as "four." JPG can easily do 5 stops, and 6 is doable, with 8 being the absolute limit.

This is not Dave's world of arbitrary mathematics. You can't just state that JPG does this and that, when it's clearly not true, and you have no basis for this.

Yeah, maybe you did. Having "less stops of tone than raw" is a little bit different than "it has 4 stops!!!"

Perhaps next time you get pulled over for speeding, the officer should say "you were going 120!" Even if you were only really going "more than 65." Why be specific, right?

This thread is a good example of why engineers should not engage in arguments with artists.

rmford wrote in post #3662124
On a serious note, though, would a test for tonal range breadth be like the test thing they do on dpreview, with the gradiated bar? I guess the best way i can think of would be to have the bar going from 0%-100% grey in segments, with each segment having a black bar above, and a white bar below. Check in PS where the furthest points apart lie that you can still see a difference between the grey and the black/white, and you have your dynamic range, no?

You are correct, this is a way to test dynamic range. However, you do not necessarily need the white and black bars, as you just need to determine the point where the signal flatlines at a minimum or a maximum.

rmford wrote in post #3662124
Expose properly, negate the problem

On a serious note, though, would a test for tonal range breadth be like the test thing they do on dpreview, with the gradiated bar? I guess the best way i can think of would be to have the bar going from 0%-100% grey in segments, with each segment having a black bar above, and a white bar below. Check in PS where the furthest points apart lie that you can still see a difference between the grey and the black/white, and you have your dynamic range, no?

Or am i thinking something completely different?

No the point is to see, using the graduated bar, where in the sequence it is not possible to differentiate the previous bar from the next adjacent bar...then you are at one limit of the number of levels of gradation; do the same at the other end of the tonal scale, and you are at the other end.

With film, there were step wedges with 0.3EV steps, to assess film response.

davesrose wrote in post #3661809
I think what has confused everyone (myself included) is that we tend to think about "dynamic range" as being only a function of the camera sensor. There is a dynamic range with the total luminosity range of the scene you are photographing, another one for the light sensitivity of the sensor recording, and another one for the bit depth of the stored image (with jpeg having less range then the sensor is capable of and RAW having more range).

Well, in many cases, it IS the sensor that limits DR. It's just not the only factor that comes into play.

I just got an XTi as a backup camera, and while it is a good performer, it's apparent (especially at ISO 400 and up) that it doesn't handle noise as well as the 5D. Noise limits or sets the exposure floor (the dark end of the exposure spectrum) regardless of how finely (8, 12, 14 bit) you divide the increments of brightness change.

Noise tends to mask detail in the darker areas of the image, especially when you try to pull exposure up a bit. With the XTi, it's more critical that you expose "to the right" to try to get the low end up above the noise as much as possible. I can rescue an underexposed 5D image and maintain shadow detail to a greater extent than I can an XTi image exposed equally (I won't dive into the XTi's tendency to expose about 1/3 stop lower than other cameras I've owned - I just use +1/3 as my "zero" on the compensation scale). That's not a criticism of the XTi so much as it's praise for the 5D - it simply has lower noise and that translates in to being able to use more of the data in the shadows of the image. It makes the useful dynamic range greater.

I'm still testing and comparing the two as time permits so that I know how the Xti will act in a given situation.

kpt4321 wrote in post #3662189
This thread is a good example of why engineers should not engage in arguments with artists.

Thankfully, I'm neither.

(although I'm probably closer to engineer, given my work and training background....)

kpt4321 wrote in post #3662189
Tom's table showed that 8 bits have a MAXIMUM dynamic range of 8 stops, and a more feasible dynamic range (while still retaining some brightness resolution within each stop) less than that.

You stated that jpg was limited to 4 stops, or "little more than 4 stops." You clearly are not a mathematician, but "less than 8" is not the same as "four." JPG can easily do 5 stops, and 6 is doable, with 8 being the absolute limit.

And you don't seem to grasp that a JPEG has 256 levels of contrast to RAW's 4092. I meant that Jpeg has more of a workable level of 4 stops....as once you get past stop 4, that's 16 tones. Obviously it could get even 5 stops if the sensor's range is less (depending on what the luminosity of the captured scene was). But it doesn't take a mathematician to know that that's low numbers for distributing tones the further down in stops you go!!!!!!

kpt4321 wrote in post #3662189
This thread is a good example of why engineers should not engage in arguments with artists.

I realize you don't give one damn about color theory, kpt4321: and being an engineer, you only look at the camera sensor and not the system....I'm still waiting to hear why Bayer sensors are like the human eye, and don't follow the ultimate color space that the image will take (RGB). It's not pasted to our own brains, so don't give me papers on the human eye. Why did Bayer make green a dominant luminance channel? Could it be his words that the green sensor is the luminance element is correct??

Anyway, I'm done with your trolling!!!

Tom W wrote in post #3662289
Well, in many cases, it IS the sensor that limits DR. It's just not the only factor that comes into play.

I just got an XTi as a backup camera, and while it is a good performer, it's apparent (especially at ISO 400 and up) that it doesn't handle noise as well as the 5D. Noise limits or sets the exposure floor (the dark end of the exposure spectrum) regardless of how finely (8, 12, 14 bit) you divide the increments of brightness change.

Righto....As I've found from a more friendly engineer on this forum.....there is dynamic range of measurement. This is a whole system, so there are different ranges for each level. The sensor is certainly THE distinguishing feature of 12 bit cameras. The 5D would have less noise at a given level because of its sensor. PQ is effected by both the sensor's saturation of contrast and noise level, as well as the processed bit depth. That's what's not mutually exclusive.

Wilt wrote in post #3662219
No the point is to see, using the graduated bar, where in the sequence it is not possible to differentiate the previous bar from the next adjacent bar...then you are at one limit of the number of levels of gradation; do the same at the other end of the tonal scale, and you are at the other end.

With film, there were step wedges with 0.3EV steps, to assess film response.

It's a good visual test, and is valid though subject to subjective interpretation (say that 3 times). DPReview uses that test (or a similar test), but it appears that Phil has only used it on JPEG images, meaning that most any camera will present the same DR, or within 1/3 stop. He does delve into RAW now, and uses a couple of different processing software products to try to get the best performance out of a camera.

BTW, I like Phil's testing. When it is suspect (such as the JPG dynamic range tests), he adds more tests or alters the procedure to make it more valid. He shows his work!

davesrose wrote in post #3662382
And you don't seem to grasp that a JPEG has 256 levels of contrast to RAW's 4092.

You're kidding me, right? This is elementary mathematics. 2^8=256, 2^12=4092. If you're still hung up here, we have a problem.

I'm still waiting to hear why Bayer sensors are like the human eye, and don't follow the ultimate color space that the image will take (RGB). It's not pasted to our own brains, so don't give me papers on the human eye. Why did Bayer make green a dominant luminance channel? Could it be his words that the green sensor is the luminance element is correct??

The green is the dominant channel due to the fact that our eyes are more sensitive to green. Maybe you missed the memo, but after we take pictures with digital cameras, we look at them with our eyes. As such, it is crucial that we capture the data in a fashion which allows us to best project it in a manner which is most effective in terms of human vision. The job of a digital camera sensor is to record image data such that it can be made into a picture to be seen by human eyes, thus the properties of the human eye ARE of consequence.

It even says this on your precious wikipedia, on the bayer filter page:

"He used twice as many green elements as red or blue to mimic the human eye's greater resolving power with green light"

What do you mean by "the green sensor is the luminance element"? All three colors on the sensor measure luminance, and the information from all three colors is used to determine both the brightness and the hue of any given pixel. the luminance information comes from all three colors, and the hue information comes from all three colors.. Yes, they are not the same ratio, that does not change this fact.

I take it upon myself to try to correct your technical misinformation which you are spreading across multiple threads, and you label me a troll. So be it.

kpt4321 wrote in post #3662566
You're kidding me, right? This is elementary mathematics. 2^8=256, 2^12=4092. If you're still hung up here, we have a problem.

And so you have a much larger workable range in f stops with RAW.....I rest my case!!!!

kpt4321 wrote in post #3662566
The green is the dominant channel due to the fact that our eyes are more sensitive to green. Maybe you missed the memo, but after we take pictures with digital cameras, we look at them with our eyes.

And you keep ignoring the fact that the captured data becomes RGB.....we are not seeing an image that has more green then red or blue. You're still completely ignoring color theory!!!

You don't seem to read do you? I have told you the meaning of the term many times.

"Bryce Bayer's patent called the green photosensors luminance-sensitive elements and the red and blue ones chrominance-sensitive elements"

And you have told me that green is not a luminance channel, which is also untrue:

"Digital camera raw files contain the pixel data from a rectangular image sensor, the modern equivalent of traditional film, usually at 12 or 14 bits per sensor bucket. The sensor is almost invariably overlaid with a so-called Bayer filter, consisting of a mosaic of red, blue and green filters in alternating rows of RG and GB. Given that three colors fit uncomfortably in a rectangular grid, green was chosen to be doubly present, since the human eye is more sensitive to it. Green also often serves as the luminance channel, and as the dominant channel for in-camera black-and-white conversions. To retrieve an image from a raw file, this mosaic of data must be converted into a full RGB image. This is known as demosaicing, but is sometimes referred to as digital development"

Yes, it doesn't change the fact that you keep on ignoring color space

kpt4321 wrote in post #3662566
I take it upon myself to try to correct your technical misinformation which you are spreading across multiple threads, and you label me a troll. So be it.

You completely ignore my arguements and keep with saying that green is just dominant because that's what our eyes are like. The final image is not green dominant....and you have even said it's wrong to label the green channel a luminance channel, as well as the green photo sensors being luminance elements. Even though the inventor himself calls the green one "luminance-sensitive elements". Since I have not seen any arguements to this fact..instead your only argument is the smoke mirror that the bayer sensor was designed like our own eye..then yes, you seem to be wanting to troll.

davesrose wrote in post #3661809
....... I think what has confused everyone (myself included) is that we tend to think about "dynamic range" as being only a function of the camera sensor. There is a dynamic range with the total luminosity range of the scene you are photographing, another one for the light sensitivity of the sensor recording, and another one for the bit depth of the stored image (with jpeg having less range then the sensor is capable of and RAW having more range).

I agree with many of the things that you have stated, however, the dynamic range is not a function of whether the recorded image is stored in 12 or 8 bits. The lower number of bits only means that the ability to distinguish between levels of luminosity will be diminished. We can define black and white levels for an eight-bit image in software to be the same as we would be for a twelve-bit image ... it's just that the eight bit image would not be able to distinguish between various levels most especially towards the dark end of luminosity.

On a completely different note, have any of you heard about the new Kodak sensor that in effect processes images in opponency similar to the L*a*b* color model? The advantage of the new design is that it is able to obtain approximately a two EV increase in sensitivity. I first saw this information in Electronic Products Magazine, a trade publication for Electronics Engineers. I may post this information in a new thread so that I do not continue to hijack this one.

davesrose wrote in post #3662651
...... "Bryce Bayer's patent called the green photosensors luminance-sensitive elements and the red and blue ones chrominance-sensitive elements" ........ Green also often serves as the luminance channel, and as the dominant channel for in-camera black-and-white conversions. To retrieve an image from a raw file, this mosaic of data must be converted into a full RGB image. ......

I believe that you might be taking the statement too literally ... it is true that most of the luminance in visible light is in the green part of the spectrum because that is where our visual acuity peaks. However, the data is not processed as a L*a*b* model so the only place where it is actually treated as luminosity data would be for B/W conversions as your quoted source says.

The modified Bayer filter that I alluded to in my previous post does use actual luminosity data from high sensitivity panchromatic (clear) sensors to produce its L*a*b* output.

bill boehme wrote in post #3662850
I may post this information in a new thread so that I do not continue to hijack this one.

I'm the one that should apologize for taking up this thread and arguing with a certain someone.....apologies for everyone else!

bill boehme wrote in post #3662850
I agree with many of the things that you have stated, however, the dynamic range is not a function of whether the recorded image is stored in 12 or 8 bits. The lower number of bits only means that the ability to distinguish between levels of luminosity will be diminished.

The main thing, I think, is that "dynamic range" is a general term....and that's where the main futile arguement is. When you get to the file size stage of the system, it's the contrast range that determines its dynamic range. So with 12bit, you have "a range" of 4092 levels of contrast, and with 8 bit you have 256. Now when we have, mainly, 12bpc cameras.....we know that sensors have a light sensitivity range approaching 4092 levels. When comparing PQ of 12bit cameras, the only difference is sensors. And a sensor's dynamic range is also dependant on how much of a luminosity range the scene has.

Because its a system, dynamic range is different at each stage....much like analog photography, which has luminosity range of scene, Light sensitivity range of film, opacity range of developed film images, and reflectance range of images on photographic papers

With digital, it's luminosity range of scene, light sensitivity of sensor, signal conditioning, processing engine, and then storage medium

Since I do a lot of 3D rendering, I also just think of the file size as it's total contrast range....so in my world it is most important DR But in photography, you still get most of your sensor information if the file format can match the amount of luminosity that the camera can capture.

bill boehme wrote in post #3662927
I believe that you might be taking the statement too literally ...

I was just arguing with someone who wants to flame me...in a previous post I mentioned how in color theory, hue plays a secondary role to value of contrast...in general terms, the green data is carrying the most luminance data: hence why Bayer termed the green sensor that, and why certain cameras just use the green channel as their histogram (since it represents the most contrast and saves time processing). You're coming up with other reasons why sensors have a dominant number of green

Oh....thinking of color theory again...one of the reasons why they chose to have extra green was so that the photo site would then have an even number of cells (and a full distribution for light capture). Having a denser amount of green sensors gives it more contrast then red or blue. So having more values of green can help make it the channel that produces less noise for the overall RGB image (since it has the same range as red and blue, but more levels).