If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Display Modes |
#381
|
|||
|
|||
Finally got to the point where no new camera holds my interest(waiting for specific offering)
On 2019-01-17 17:50, Eric Stevens wrote:
On Thu, 17 Jan 2019 10:34:05 -0500, nospam it's basic sampling theory, something which you clearly do not understand at all. What is the frequency of the signal being sampled? That's the point. The sample of any pixel is a discrete sample for the purpose of dynamic range. There is no time domain here. If you were looking at image resolution, then you would find that spatial sampling across a number of pixels can be largely compared to sampling a time domain signal from a math POV. But that is not about DR or how much information can fit in a sample of so many bits. -- "2/3 of Donald Trump's wives were immigrants. Proof that we need immigrants to do jobs that most Americans wouldn't do." - unknown protester |
#382
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
On Fri, 18 Jan 2019 17:04:58 -0500, Alan Browne
wrote: On 2019-01-17 17:31, Eric Stevens wrote: On Thu, 17 Jan 2019 11:19:16 -0500, Alan Browne wrote: On 2019-01-17 09:22, Alan Browne wrote: The link you pointed to was referring to quantization noise in the frequency domain, not the amplitude which is the issue at hand here. Ooops... I should have said "time domain". Nevertheless, the principle remains, the digitization of a non-identical source signal can only be an approximation of the original signal. The differences/errors manifest themselves as noise. Nothing to do with DR however... esp. not getting more than there is. I agree that it has nothing to do with DR. As far as your second comment, the argument has been about whether or not the ADC can receive a wider DR than it can handle. The answer is yes, of course it can, but all the ADC can output is inherent in it's bit width. This is where all the argument has arisen. The DR inherent in the ADC is not the same as the DR inherent in the sensor. It is the latter which is tested and reported upon by DxO. You may think that is a waste of time and even misleading. But it isn't. The ADC doesn't create the image: the sensor does and the ADC only handles the image at the start of a long and convoluted pipeline which ends up on the much narrower DR of your screen. -- Regards, Eric Stevens |
#383
|
|||
|
|||
Finally got to the point where no new camera holds my interest(waiting for specific offering)
On 1/18/2019 6:45 PM, Eric Stevens wrote:
On Fri, 18 Jan 2019 17:04:58 -0500, Alan Browne wrote: On 2019-01-17 17:31, Eric Stevens wrote: On Thu, 17 Jan 2019 11:19:16 -0500, Alan Browne wrote: On 2019-01-17 09:22, Alan Browne wrote: The link you pointed to was referring to quantization noise in the frequency domain, not the amplitude which is the issue at hand here. Ooops... I should have said "time domain". Nevertheless, the principle remains, the digitization of a non-identical source signal can only be an approximation of the original signal. The differences/errors manifest themselves as noise. Nothing to do with DR however... esp. not getting more than there is. I agree that it has nothing to do with DR. As far as your second comment, the argument has been about whether or not the ADC can receive a wider DR than it can handle. The answer is yes, of course it can, but all the ADC can output is inherent in it's bit width. This is where all the argument has arisen. The DR inherent in the ADC is not the same as the DR inherent in the sensor. It is the latter which is tested and reported upon by DxO. You may think that is a waste of time and even misleading. But it isn't. The ADC doesn't create the image: the sensor does and the ADC only handles the image at the start of a long and convoluted pipeline which ends up on the much narrower DR of your screen. IMHO the real question was if/how one might see beyond the firewall of the ADC to glean this additional DR information. Thus far the answer has been a big NO. [YMMV] -- == Later... Ron C -- |
#384
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
On Fri, 18 Jan 2019 17:02:36 -0500, Alan Browne
wrote: On 2019-01-17 17:28, Eric Stevens wrote: On Thu, 17 Jan 2019 09:22:21 -0500, Alan Browne wrote: On 2019-01-16 23:43, Eric Stevens wrote: On Tue, 15 Jan 2019 19:42:49 -0500, Alan Browne wrote: On 2019-01-15 18:29, Eric Stevens wrote: On Mon, 14 Jan 2019 12:34:42 -0500, Alan Browne wrote: On 2019-01-13 21:07, Eric Stevens wrote: On Sun, 13 Jan 2019 10:06:13 -0500, Alan Browne wrote: On 2019-01-12 21:15, Eric Stevens wrote: On Sat, 12 Jan 2019 09:25:23 -0500, Alan Browne wrote: On 2019-01-11 18:18, Eric Stevens wrote: The recorded output of the ADC is limited by the capabilities of the ADC. But these have no effect on the capabilities of the sensor. If If there is no way to encode the information, then that is the mootest of moot points. That may well be but, as I have several times said, it is possible to scale the dynamic range of the sensor to fit the narrower dynamic range of the ADC. To which I've replied numberous times. In a nutshell, you're trading one form of noise for another. The 'scaling' is done during the conversion of analog to digital in the ADC and involves no more noise than is inherent in any analog to digital conversion. I've pointed out quantization noise to you several times. Ignore it. You ignore everything else. You always get quantization noise when you digitize. That's what I meant by 'inherent'. No. A sample is a sample. It contains signal and _sampling_ source noise (the noise of the thing being sampled and the noise of the ADC), but not quantization noise. In this case we are not considering sampling. We are dealing with digitizing of a static (as in remains constant) charge in the pixel well which has to be digitized as a proportion of the maximum charge when the cell is filled. Quantization noise is an artifact of data manipulation (such as scaling) and is completely unavoidable when doing such operations. Its also inevitable when digitizing an analog signal. Which you clearly don't understand. The link you pointed to was referring to quantization noise in the frequency domain, not the amplitude which is the issue at hand here. I was clearly referring to noise introduced by scaling or manipulation (amplitude). But you will continue to hunt for anything that violates the notion that there is more information in 14 bits than there is in 14 bits. There isn't. There can't be. Frequency or amplitude: the principle is the same. The difference here is very simple, however. Here (photography) you are sampling an static analog signal that has noise as part of the overall quantity. That noise is just random numbers in the lower bits... No quantizuation noise at all. Yes. That's analog noise, if there is such a term. And that noise will later be digitised. Then you claim something special is a happening. No matter what that magic is (compression, dilation, re-mapping (LUT)), etc. will introduce quantization noise when the information is extracted. Because: 14 bits. There will always be quantization noise if only for the reason that e.g. analog 10.5 in can on be digitized as 10 or 11. Either way there is an error of -0.5 or +0.5. This is _not_ related to time domain sampling quantization noise which is a different beast (which in any case should be taken care of by appropriate attention to Nyquist, but that's another story that belongs to sampling a signal over time...). I don't think Nyqust really comes into it as we are dealing with the digitization of static values. I have never argued that "there is more information in 14 bits than there is in 14 bits". As you say, there can't be. What I have said from the beginning is that an analog signal of _any_dynamic_range_ can be coded in 14 bits. There is no reqirement imposed on the DR of the source device. The dynamic range is simply represented by bits of sample depth. That's the dynamic range of the digitized signal. That's the simplest way to look at it. If there was some sort of compression before that point then the benefit of it would be lost to quantization noise in the representation in 14 bits and then "re-conversion" to express it as more bits. The ADC doesn't know anything about the sensor's DR as measured in photons. All it knows is that is being asked to digitize a signal which may be anywhere between 0V and 3.3V. It is this which is chopped up into 12 bits or 14 bits or whatever. There is another thing which has barely been touched upon. We don't actually how Nikon (in this case) sets about coding the information from the sensor when it packs it into the bit stream. In fact it makes no conventional sense until it is unpacked by a raw decoder. We don't even know that the sensor's ADC is 14 bits. We have just assumed that it is. But does it just convert itself to a 14 bit ADC when the menu is set accordingly? I very much doubt it. But many cameras these days have DR greater than 12 while only using 12 bit encoding. All we know is that after a D800 etc has finished dealing with things like demosaicing, tone mapping, white balance, denoising, sharpening, compression it outputs a file which somewhere along the way has been processed with a bit depth of 14. It remains a fairy tale, IOW: TANSTAAFL. I have said this in so many ways so many times that I'm not going to try and repeat it any more. That's a good thing. Because it is as wrong after n repetitions as it was after the first. Yes, but just how wrong is that? -- Regards, Eric Stevens |
#385
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
On Fri, 18 Jan 2019 17:10:41 -0500, Alan Browne
wrote: On 2019-01-17 17:50, Eric Stevens wrote: On Thu, 17 Jan 2019 10:34:05 -0500, nospam it's basic sampling theory, something which you clearly do not understand at all. What is the frequency of the signal being sampled? That's the point. The sample of any pixel is a discrete sample for the purpose of dynamic range. There is no time domain here. Exactly. Which is why I asked and no doubt why ****** evaded. If you were looking at image resolution, then you would find that spatial sampling across a number of pixels can be largely compared to sampling a time domain signal from a math POV. Yes, but it depends upon the method of sampling the pixels. But even so, the actual sampling is one pixel at a time. But that is not about DR or how much information can fit in a sample of so many bits. Quite, which is what I was hoping ****** would commit himself to pointing out. -- Regards, Eric Stevens |
#386
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
In article , Eric Stevens
wrote: There is another thing which has barely been touched upon. We don't actually how Nikon (in this case) sets about coding the information from the sensor when it packs it into the bit stream. In fact it makes no conventional sense until it is unpacked by a raw decoder. nonsense. of course we do. shoot some test images and look at the data. or take the lazy approach and look at what others have found. We don't even know that the sensor's ADC is 14 bits. We have just assumed that it is. are you claiming nikon is lying when they say it's 14 bits? it's also not that hard to verify. But does it just convert itself to a 14 bit ADC when the menu is set accordingly? I very much doubt it. what else would it do? But many cameras these days have DR greater than 12 while only using 12 bit encoding. no they don't. if they have a 12 bit adc, they're limited to 12 stops, thus the option for 14 bit. All we know is that after a D800 etc has finished dealing with things like demosaicing, tone mapping, white balance, denoising, sharpening, compression it outputs a file which somewhere along the way has been processed with a bit depth of 14. which has a theoretical maximum of 14 stops, real world less than that. It remains a fairy tale, IOW: TANSTAAFL. I have said this in so many ways so many times that I'm not going to try and repeat it any more. That's a good thing. Because it is as wrong after n repetitions as it was after the first. Yes, but just how wrong is that? very. |
#387
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
In article , Eric Stevens
wrote: it's basic sampling theory, something which you clearly do not understand at all. What is the frequency of the signal being sampled? That's the point. The sample of any pixel is a discrete sample for the purpose of dynamic range. There is no time domain here. Exactly. Which is why I asked and no doubt why ****** evaded. i didn't evade anything. i simply pointed out that you don't understand the topic. |
#388
|
|||
|
|||
Finally got to the point where no new camera holds my interest (waiting for specific offering)
On Fri, 18 Jan 2019 19:37:31 -0500, Ron C wrote:
On 1/18/2019 6:45 PM, Eric Stevens wrote: On Fri, 18 Jan 2019 17:04:58 -0500, Alan Browne wrote: On 2019-01-17 17:31, Eric Stevens wrote: On Thu, 17 Jan 2019 11:19:16 -0500, Alan Browne wrote: On 2019-01-17 09:22, Alan Browne wrote: The link you pointed to was referring to quantization noise in the frequency domain, not the amplitude which is the issue at hand here. Ooops... I should have said "time domain". Nevertheless, the principle remains, the digitization of a non-identical source signal can only be an approximation of the original signal. The differences/errors manifest themselves as noise. Nothing to do with DR however... esp. not getting more than there is. I agree that it has nothing to do with DR. As far as your second comment, the argument has been about whether or not the ADC can receive a wider DR than it can handle. The answer is yes, of course it can, but all the ADC can output is inherent in it's bit width. This is where all the argument has arisen. The DR inherent in the ADC is not the same as the DR inherent in the sensor. It is the latter which is tested and reported upon by DxO. You may think that is a waste of time and even misleading. But it isn't. The ADC doesn't create the image: the sensor does and the ADC only handles the image at the start of a long and convoluted pipeline which ends up on the much narrower DR of your screen. IMHO the real question was if/how one might see beyond the firewall of the ADC to glean this additional DR information. Thus far the answer has been a big NO. [YMMV] I'm not aware that anyone has previously asked that question. I think the answer is that it could be done by the decoder. Whether in the raw file created by the camera or whether embedded in the raw decoder for each nikon camera, there could be a factor to restore the original DR subject only to the losses imposed by 14-bit digitizing and the noise created by swapping the data from one format to another. In any case these might be of interest https://www.dpreview.com/forums/post/15353950 https://www.dpreview.com/forums/post/15363777 https://forum.nikonrumors.com/discus...ge-of-printers -- Regards, Eric Stevens |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Finally got to the point where no new camera holds my interest (waiting for specific offering) | Alfred Molon[_4_] | Digital Photography | 2 | December 24th 18 02:37 PM |
Please, tell me Zeiss's offering to the camera world won't be areskinned SONY!! | Neil[_9_] | Digital Photography | 1 | August 27th 18 01:00 PM |
Need a camera with specific features: | Gary Smiley | Digital Photography | 1 | May 22nd 06 02:31 AM |
Canon Offering $600+ Rebate on Digital Camera Equipment (3x Rebate Offers) | Mark | Digital Photography | 6 | November 4th 04 10:27 AM |