This article should be read by everybody

Started Dec 12, 2013 | Discussions
K E Hoffman
K E Hoffman Veteran Member • Posts: 5,103
Re: This article should be read by everybody
1

cosmonaut wrote:

There are a few out there that will argue DXO marks are useless imformation and has little to do with image quality in the real world. I have noticed those people generally shoot cameras with low scores. I almost never hear a photographer argue the point when they shoot with a camera that scores high.

I have always found DXO marks accurate from what cameras I have used in the past.

Or they are looking for a human truth that can't be boiled down to a single number?

I own the best rated Sony Current Model Sony APC DSLR on the list.. and I still question their numbers. I think they fail to tell the whole story.  Just like I question the DPR final scores.. but at least that comes with pages of thinking I can use to find what matters to me.

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony a6500 Sony E 18-200mm F3.5-6.3 OSS Sony E 50mm F1.8 OSS +3 more
artlmntl Senior Member • Posts: 1,804
Re: Proof is harder to obtain than you think

tbcass wrote:

Supposedly the A55 and A65/77 use the same sensor technology while the A57 sensor was "tweaked" for better high iso. The problem with that is DXO scores the A55 sensor a touch better at high iso than the A57 (although the difference is insignificant). There must be a margin of error in the DXO tests. Somebody else said 1/3 - 1/2 stop MOA.

I suspect Sony's "tweak" involved adding a little NR to the A57 RAW file at high iso. I think the A77 files that DPR posted are underexposed - They are screwed up images. KEH has been using those images in his extensive per pixel argument, and they don't prove anything.

I agree with Allan Oleson that the A65 files are better, seemingly more representative of what the Sony 24MP sensor is capable of doing. The overall look of the files makes me think that DPR is doing the A77 a big disservice. I also think the character of noise from the 24MP sensor, once you filter out the chroma noise, is actually preferable to the character of noise in the 16mp sensor, at least with the A57 files.

I think it's sort of unavoidable that the relationship between detail and noise that goes on at the per-pixel level has an influence on the look of the whole image, just as printing has an influence on the look of the image. But people know, especially people who have shot high-iso film, that the capabilities of current digital cameras are quite amazing.

It appears to me that there is no real reason to shy away from the A65/A77 because of high iso noise, but people who absolutely must have better high iso performance than APS-C can provide will have to write the bigger checks and carry the heavier loads for FF gear and brighter lenses.

-- hide signature --

Hunter

Robsphoto
Robsphoto Senior Member • Posts: 1,219
Noise doesn't happen at the SENSOR level, it happens PIXEL by pixel
1

SUMMARY:

Subdividing a sensor using similar technology and design DOES increase NOISE.. Just a fact. Smart people each year find ways to reduce noise so the smaller pixel has a better Signal to Noise Ratio and the noise cost of More MP is reduced..

Does the noise price undo the value of more MP?... in all but the extreme cases.. I don't think so. I am less worried about 1:1 pixel sensor noise than I used to be.. because scaling does help and PP software gets better. But you can't make up facts.. buy ignoring actual changes that happen in a sensor when it is subdivided and thinking of it as a theoretical collective when noise increase happens per pixel.

Thanks very much for a very interesting post in which you say that:

“The key is NOISE does not happen at the SENSOR level. It happens PIXEL by pixel.”

I would be interested in your opinion about this statement from Sony about the relatively high pixel count of the Sony RX100:

Question to Sony:

"Why did you want a 20.2-megapixel sensor? I heard that this incredibly high pixel count would negatively impact noise levels, thereby decreasing image quality at high ISO settings.

Answer from Sony:

Ueda (Image Quality Design)

"It’s true that increasing pixel count increases noise. But since we manufacture our own sensors, we can easily tweak sensor specs to suit specific needs. This allowed us to craft a totally new sensor that delivers superbly detailed images with low noise. For high-sensitivity shooting we managed to reduce noise levels below those of existing Cyber-shots by combining technologies from Cyber-shot and α Series. As a result, we can shoot at up to ISO 6400 for normal photos and up to ISO 25600 when using Multi Frame NR."

http://www.sony.com.sg/microsite/cybershot/rx100/pdf/performance.pdf

Also in the above statement, Kaimi (Product Design) said this:

"Some compact cameras keep noise down with lower pixel counts. But there are users who really want super-fine images, which require a high pixel count. But if the higher pixel count leads to unacceptable noise levels that the noise reduction system can’t reduce without affecting image quality, it is essentially self-defeating. Our new sensor achieves both super-fine images and low noise so even distant landscapes shot at telephoto focal lengths show great detail. People are going to love this compact camera for travel photography."

This question is to K E Hoffman and asks, in light of the information given in your very informative posts to this thread, do you think Ueda’s statement that “it’s true that increasing pixel count increases noise” is reasonable from a technical point of view? If so, do you think that Ueda was referring to noise at the pixel level, or noise that would be visible at the IMAGE level (if Sony hadn’t “managed to reduce noise levels below those of existing Cyber-shots”)?

Incidentally, I have started to construct a page on my web site that includes expert opinion from many different web sites so that readers can make up their own minds about these issues when they are deciding which camera best suits their needs:

http://www.robsphotography.co.nz/high-iso-low-light.html

Cheers

Rob

K E Hoffman
K E Hoffman Veteran Member • Posts: 5,103
Thanks - Sensor Designer Sony Confirms my Opinion
1

Robsphoto wrote:

SUMMARY:

Subdividing a sensor using similar technology and design DOES increase NOISE.. Just a fact. Smart people each year find ways to reduce noise so the smaller pixel has a better Signal to Noise Ratio and the noise cost of More MP is reduced..

Does the noise price undo the value of more MP?... in all but the extreme cases.. I don't think so. I am less worried about 1:1 pixel sensor noise than I used to be.. because scaling does help and PP software gets better. But you can't make up facts.. buy ignoring actual changes that happen in a sensor when it is subdivided and thinking of it as a theoretical collective when noise increase happens per pixel.

Thanks very much for a very interesting post in which you say that:

“The key is NOISE does not happen at the SENSOR level. It happens PIXEL by pixel.”

I would be interested in your opinion about this statement from Sony about the relatively high pixel count of the Sony RX100:

Question to Sony:

"Why did you want a 20.2-megapixel sensor? I heard that this incredibly high pixel count would negatively impact noise levels, thereby decreasing image quality at high ISO settings.

Answer from Sony:

Ueda (Image Quality Design)

"It’s true that increasing pixel count increases noise. But since we manufacture our own sensors, we can easily tweak sensor specs to suit specific needs. This allowed us to craft a totally new sensor that delivers superbly detailed images with low noise. For high-sensitivity shooting we managed to reduce noise levels below those of existing Cyber-shots by combining technologies from Cyber-shot and α Series. As a result, we can shoot at up to ISO 6400 for normal photos and up to ISO 25600 when using Multi Frame NR."

http://www.sony.com.sg/microsite/cybershot/rx100/pdf/performance.pdf

Also in the above statement, Kaimi (Product Design) said this:

"Some compact cameras keep noise down with lower pixel counts. But there are users who really want super-fine images, which require a high pixel count. But if the higher pixel count leads to unacceptable noise levels that the noise reduction system can’t reduce without affecting image quality, it is essentially self-defeating. Our new sensor achieves both super-fine images and low noise so even distant landscapes shot at telephoto focal lengths show great detail. People are going to love this compact camera for travel photography."

This question is to K E Hoffman and asks, in light of the information given in your very informative posts to this thread, do you think Ueda’s statement that “it’s true that increasing pixel count increases noise” is reasonable from a technical point of view? If so, do you think that Ueda was referring to noise at the pixel level, or noise that would be visible at the IMAGE level (if Sony hadn’t “managed to reduce noise levels below those of existing Cyber-shots”)?

Incidentally, I have started to construct a page on my web site that includes expert opinion from many different web sites so that readers can make up their own minds about these issues when they are deciding which camera best suits their needs:

http://www.robsphotography.co.nz/high-iso-low-light.html

Cheers

Rob

My opinion is that he said exactly what I have been saying. Increased Pixel count in same size sensor increases noise.. but you can improve the design of the sensor (that includes pixel level improvements and support electronics improvements to reduce that noise)

Since this comes from an member of a sensor design team, not an academic. And it is released with marketing team approval who would love to say there is no noise cost to more MP. It tells me the MYTH is that MORE MP doesn't affect noise.

IT does, he says so as one of the leading sensor makers on the planet. He is also very proud that they find ways to improve the sensor response and reduce noise.

But also as I say in several of my posts.. the noise coverage in reviews is overblown and there is no current DSLR where you would not apply some NR to the RAW PP and they always compare RAW with no NR.

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony a6500 Sony E 18-200mm F3.5-6.3 OSS Sony E 50mm F1.8 OSS +3 more
jonas ar Contributing Member • Posts: 606
Re: This article should be read by everybody

K E Hoffman wrote:

cosmonaut wrote:

There are a few out there that will argue DXO marks are useless imformation and has little to do with image quality in the real world. I have noticed those people generally shoot cameras with low scores. I almost never hear a photographer argue the point when they shoot with a camera that scores high.

I have always found DXO marks accurate from what cameras I have used in the past.

Or they are looking for a human truth that can't be boiled down to a single number?

I own the best rated Sony Current Model Sony APC DSLR on the list.. and I still question their numbers. I think they fail to tell the whole story. Just like I question the DPR final scores.. but at least that comes with pages of thinking I can use to find what matters to me.

Well, they do show the underlying data if you take the time. I agree the composit numbers are pretty useless.

tbcass
OP tbcass Forum Pro • Posts: 41,258
Re: Noise doesn't happen at the SENSOR level, it happens PIXEL by pixel

There was a discussion about that a little while ago. It was also pointed out that another Sony report from a couple years ago said just the opposite, that increasing pixel density does not increase noise. Those guys you quoted are not engineers but were referred to as "designers" who probably have more to do with deciding how the camera is built and looks, not the electronics.

-- hide signature --

Tom
Look at the picture, not the pixels
------------
Miss use of the ability to do 100% pixel peeping is the bane of digital photography because it causes people to fret over inconsequential issues.
http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/

 tbcass's gear list:tbcass's gear list
Sony RX100 Sony RX10 IV Sony a99 II Tamron SP AF 90mm F/2.8 Di Macro Sony DT 35mm F1.8 SAM +10 more
K E Hoffman
K E Hoffman Veteran Member • Posts: 5,103
Re: Noise doesn't happen at the SENSOR level, it happens PIXEL by pixel
1

tbcass wrote:

There was a discussion about that a little while ago. It was also pointed out that another Sony report from a couple years ago said just the opposite, that increasing pixel density does not increase noise. Those guys you quoted are not engineers but were referred to as "designers" who probably have more to do with deciding how the camera is built and looks, not the electronics.

His title in the interview is (Image Quality Designer) So no he is not working on the case design of the camera.

can't tell if its the same person..

but there is a Kihachiro Ueda listed at co-inventor on 4 Sony Patents since 1996.. that are in electronics and one having to do with coloring of transparent films with high transparency "thin plastics" not camera film.

Just because the facts are the facts doesn't mean Sony has not done a good job getting more out of the same sized sensor over the last 8 years.  If you look at the Older Sony A580 sensor vs the A77 in the silly RAW with no NR view.. The A77 has more noise at just about every level.. But if you go t JPG the sometimes maligned Sony A77 JPG engine on the camera. the A77 finds all the detail of the A77 .. so the camera uses not only sensor design.. but the improvements in RAW processing to get better images.

-- hide signature --

Tom
Look at the picture, not the pixels
------------
Miss use of the ability to do 100% pixel peeping is the bane of digital photography because it causes people to fret over inconsequential issues.
http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony a6500 Sony E 18-200mm F3.5-6.3 OSS Sony E 50mm F1.8 OSS +3 more
Robsphoto
Robsphoto Senior Member • Posts: 1,219
Sony's 16.41mp phone sensor and its very small pixels: Apple increases pixel size on the iPhone 5S

tbcass wrote:

There was a discussion about that a little while ago. It was also pointed out that another Sony report from a couple years ago said just the opposite, that increasing pixel density does not increase noise. Those guys you quoted are not engineers but were referred to as "designers" who probably have more to do with deciding how the camera is built and looks, not the electronics.

Below is a quote from my web page about the report that I think Tom is referring to. The report said there is a "misconception that reduced pixel size causes deterioration in characteristics".

"On 30 August 2011, Yasuhiro Ueda of Sony's Image Sensor Business Division, Semi Conductor Business Group, issued a semiconductor briefing "Image Sensor Business". As part of this briefing, a diagram was produced titled "Misconception that reduced pixel size causes deterioration in characteristics". This diagram has sometimes been reproduced on forums to support the view that reducing the pixel size does not necessarily lead to a noticeable deterioration in image quality when the image is viewed at the image level."

So, for Tom to say that this statement said that "increasing pixel density does not increase noise" is a bit misleading because "noise" is not actually mentioned in Sony's choice of words! In previous threads the diagram in the statement has been referred to as what happens only when the images are viewed at the image level.

In any event, the above statement is referring to a back-illuminated 16.41mp CMOS sensor that Sony developed for use in PHONES. The image size is 4672 pixels x 3512 pixels and the pixel pitch is just 1.12 microns (with a sensor width of only about 5.23mm). If we could see a range of actual full-sized high ISO images from cameras using these sensors, this may give us a better appreciation of the quality of these images when taken in low light.

Note that the megapixel race in phone cameras is now being "reversed" by Apple as discussed here.

Anyway here is another extract from an article by Sony engineers about the Exmor CMOS imaging sensor developed by Sony for use in digital cameras:

"To achieve higher pixel counts, pixels must be made smaller. However, sensitivity is generally proportional to pixel area and will decline if pixel size is reduced. By combining its knowledge of the technology developed for CCDs, Sony has been able to compensate for the sensitivity loss caused by reductions in pixel size. The future goal is to combine higher speeds with higher resolutions while compensating for reductions in sensitivity resulting from reductions in pixel size. This will be achieved using device process technology, circuit technology and image processing technology."

Cheers

Rob

K E Hoffman
K E Hoffman Veteran Member • Posts: 5,103
Re: Sony's 16.41mp phone sensor and its small pixels: Apple increases pixel size on the iPhone 5S

Robsphoto wrote:

tbcass wrote:

There was a discussion about that a little while ago. It was also pointed out that another Sony report from a couple years ago said just the opposite, that increasing pixel density does not increase noise. Those guys you quoted are not engineers but were referred to as "designers" who probably have more to do with deciding how the camera is built and looks, not the electronics.

Below is a quote from my web page about the report that I think Tom is referring to. The report said there is a "misconception that reduced pixel size causes deterioration in characteristics".

"On 30 August 2011, Yasuhiro Ueda of Sony's Image Sensor Business Division, Semi Conductor Business Group, issued a semiconductor briefing "Image Sensor Business". As part of this briefing, a diagram was produced titled "Misconception that reduced pixel size causes deterioration in characteristics". This diagram has sometimes been reproduced on forums to support the view that reducing the pixel size does not necessarily lead to a noticeable deterioration in image quality when the image is viewed at the image level."

So, for Tom to say that this statement said that "increasing pixel density does not increase noise" is a bit misleading because "noise" is not actually mentioned in Sony's choice of words! In previous threads the diagram in the statement has been referred to as what happens only when the images are viewed at the image level.

In any event, the above statement is referring to a back-illuminated 16.41mp CMOS sensor that Sony developed for use in PHONES. The image size is 4672 pixels x 3512 pixels and the pixel pitch is just 1.12 microns (with a sensor width of only about 5.23mm). If we could see a range of actual full-sized high ISO images from cameras using these sensors, this may give us a better appreciation of the quality of these images when taken in low light.

Note that the megapixel race in phone cameras is now being "reversed" by Apple as discussed here.

Anyway here is another extract from an article by Sony engineers about the Exmor CMOS imaging sensor developed by Sony for use in digital cameras:

"To achieve higher pixel counts, pixels must be made smaller. However, sensitivity is generally proportional to pixel area and will decline if pixel size is reduced. By combining its knowledge of the technology developed for CCDs, Sony has been able to compensate for the sensitivity loss caused by reductions in pixel size. The future goal is to combine higher speeds with higher resolutions while compensating for reductions in sensitivity resulting from reductions in pixel size. This will be achieved using device process technology, circuit technology and image processing technology."

Cheers

Rob

And welcome to what The marketing Team put in a PPT slide for a VP VS. what the Engineers say about the work they do.

Please note in slide #10 They NEVER use the word Noise.. What is a "characteristic" I don't know do you? I do know the attorneys were involved. Look at slide #14!!! I deal with this kind of thinking in my work all the time..

Then you gave us a statement from an Engineering team saying when you make the pixels smaller you reduce sensitivity.. so we create solutions for that. Reduced sensitivity means less signal, which means lower SNR.. so you look at ways to reduce noise.. Like I have been saying..

Smaller Pixels less signal per pixel.. when you do that you need to reduce the base noise that does not get reduced with the smaller size..

Physics and Electronics have truths.. Small pixels for no cost in noise AND / OR RD and better tech is a fantasy.

-- hide signature --

K.E.H. >> Shooting between raindrops in WA<<

 K E Hoffman's gear list:K E Hoffman's gear list
Canon EOS 450D Nikon 1 J1 Sony a6500 Sony E 18-200mm F3.5-6.3 OSS Sony E 50mm F1.8 OSS +3 more
tbcass
OP tbcass Forum Pro • Posts: 41,258
Re: Sony's 16.41mp phone sensor and its small pixels: Apple increases pixel size on the iPhone 5S

K E Hoffman wrote:

Physics and Electronics have truths.. Small pixels for no cost in noise AND / OR RD and better tech is a fantasy.

Except that a sensor with twice the pixel density has twice as many pixels. The article I referenced at the beginning showed that the signal to noise ratio of 2 small pixels combined is the same as one pixel twice the size. This is why the total noise produced depends on the size of the sensor not the number of pixels. Thus the concept of downsizing a higher resolution sensor to match the noise of a lower resolution sensor at high iso for 100% comparison.

While a 24mp sensor might have more pixel level noise at iso100-800 than a 16mp sensor which has more noise than a 10mp sensor (etc) the noise at lower iso is below the S/N threshold where noise becomes a problem. It's at iso 1600 and above the high resolution sensor loses it's advantage and things become equal.

If a 16mp sensor truly has better equalized S/N ratio than a 24mp sensor why stop there? Why not 8mp for even less noise or 4 mp? It's because other than filesize there is no drawback to more MP (up to the point where the chip wiring becomes a limitation). Some suggest that with existing tech 24mp is the the high limit for APS-C thus is the ideal pixel count for best IQ.

-- hide signature --

Tom
Look at the picture, not the pixels
------------
Miss use of the ability to do 100% pixel peeping is the bane of digital photography because it causes people to fret over inconsequential issues.
http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/

 tbcass's gear list:tbcass's gear list
Sony RX100 Sony RX10 IV Sony a99 II Tamron SP AF 90mm F/2.8 Di Macro Sony DT 35mm F1.8 SAM +10 more
brian14478
brian14478 Veteran Member • Posts: 3,703
Re: This article should be read by everybody

Tom
Look at the picture, not the pixels

Ok enough people have read the article....Whether or not it corresponds to real life results, that is with photographers,- not a chemist/physicist/psychic or reviewer to conclude.

As you write..'look at the picture, not the pixels..well lets see some a77 shots.  They could be at 50iso,doesn't matter to me... Show everyone why you like the a77- not why everyone should embrace 'mathematical proofs'.

The a65 is still a better overall camera, the a77 only has lens adjustment features because its focusing system sucks monkey bal*s.  The a55 makes the a77 look like an overworked piece of nonsense too. hahahah-Just kidding on this paragraph.

*If you drop sharpness in jpeg settings, the a65 has exactly the same noise performance perhaps better than the a77's(with jpegs). It has higher resolution at default and lowering sharpness doesn't make an image soft at all.

Check the a65 and a77 raw files here at 1600/3200/6400 iso.-They are exactly the same.

a55/a65/a77/a57/a58a100/a35/a99/a37/a580....all nice cameras. Thanks-have a great safe holiday season and i am looking forward to some shots from your new a77.-brian

lhkjacky Senior Member • Posts: 2,859
Re: Sony's 16.41mp phone sensor and its small pixels: Apple increases pixel size on the iPhone 5S

A77 (24Mp) at ISO100 Crop (1480x1480) (File Size:1.32MB)

A77 (24Mp downsized to 6Mp by PS) at ISO100 Crop (740x740) (File Size:486KB)

A77 (Upsize to 24Mp from 6Mp File Above by PS) at ISO100 Crop (1480x1480) (File Size:1.0MB)

Compare 24Mp vs 6Mp Upsized Image at ISO100 Crop (1480x1480) (File Size: 1.32MB vs 1.0MB)

 lhkjacky's gear list:lhkjacky's gear list
Sony a77 II Sony DT 16-50mm F2.8 SSM Tamron SP 70-200mm F/2.8 Di VC USD Tamron SP 150-600mm F/5-6.3 Di VC USD Ricoh GR +6 more
123Mike Veteran Member • Posts: 4,643
I disagree, noisy pixels overtaking correct pixels...

So long as the damage from noise does not screw up the color accuracy of a given grainy area, things will work out ok through de-noising and/or scaling down. But the problem is that when the line is crossed and the number of noisy incorrect pixels are too high, there is no fixing it anymore. The A77 and A65 crosses that line earlier than the A57 does. For that reason, the A57 will make a better shot in darker situations at highish ISO, like 1600 and 800 even.

I've had an A65 and have an A57, and this is what I observed.

Second, there is also damage to things to fine lines, edges, that become squiggly earlier.

And as for my confirmation bias, what satisfy me having passed up the A77, is that video is limited to ISO 1600. Had it gone to ISO 3200, the results would have been atrocious, because of the line skipping mechanism they're using to scale to HD resolution. Only now, are cameras beginning to use proper scaling algorithms fixing that problem. Had the A77 considered every sensor pixel for video, then it would probably be ok to ISO 6400. That's all video though. For photos, I'm afraid the problem is when the noisy pixels are overtaking correct pixels, in a nutshell.

lhkjacky Senior Member • Posts: 2,859
Re: Sony's 16.41mp phone sensor and its small pixels: Apple increases pixel size on the iPhone 5S

lhkjacky wrote:

A77 (24Mp) at ISO100 Crop (1480x1480) (File Size:1.32MB)

A77 (24Mp downsized to 6Mp by PS) at ISO100 Crop (740x740) (File Size:486KB)

A77 (Upsize to 24Mp from 6Mp File Above by PS) at ISO100 Crop (1480x1480) (File Size:1.0MB)

Compare 24Mp vs 6Mp Upsized Image at ISO100 Crop (1480x1480) (File Size: 1.32MB vs 1.0MB)

The difference between the 24Mp crop vs the 6Mp Upsized image crop is minimal.

Higher resolution sensor will give a larger file size, but it is not necessary show more detail.

Noise reduction is applied at base ISO and it smooth out fine detail.

A lower resolution sensor with less noise, can provide as good as a higher resolution sensor, because it require less noise reduction.

...

If A77 can shoot at 6Mp, it can shoot 4~5 times longer before buffer full.

eg. 7sec, 85 RAW+Jpg at 12fps, Infinity Fine jpg at 12fps, etc.

 lhkjacky's gear list:lhkjacky's gear list
Sony a77 II Sony DT 16-50mm F2.8 SSM Tamron SP 70-200mm F/2.8 Di VC USD Tamron SP 150-600mm F/5-6.3 Di VC USD Ricoh GR +6 more
Keyboard shortcuts:
FForum MMy threads