cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

M-JPEG vs L-JPEG for high ISO

sknaumov
Apprentice

Hi.

 

I tried to use M-size and S-size JPEG on my 80D for low light and haven't noticed any difference compared to L-size JPEG. I thought that by increasing "virtual" pixel size I will have cleaner images - as on full-frame bodies that have greater real pixel size.

 

The question is very actual for me as I want to upgrade to 90D but do not need these extra megaixels but good photos at high ISO. When I shoot horse show-jumping starts I notice that even on my FullHD monitor the photo doesn't look sharp because of noise, especially some darker areas like face (compared, say, with white horse), so 32MP has no value for me if I get mediocre FullHD results...

 

At the same time I do not want to work with every photo, buy additional programs to remove noise and downscale - are there some settings to improve in-camera processing?

 

Original question: https://photo.stackexchange.com/questions/112204/does-using-m-or-s-size-jpeg-helps-with-lowering-noi...

 

Thanks,

Sergey.

1 ACCEPTED SOLUTION

Combining photosites to get an 8 MP sensor from a 32 MP array provides a result quite similar to downsampling but does NOT provide the same noise figure as 8 million discrete photosites covering the same size sensor.

 

Downsampling works to reduce noise because the noise generated in the photosite itself AND the low noise amplifier used to increase sensor output to a usable level has a largely random distribution.  By downsampling, you are combining the noise and signal output between multiple closely spaced photosites and in best case the largely correlated signal algebraically sums and so does the noise but because the phase relationship of the noise component is randomly distributed much of it cancels resulting in less noise in a downsampled image.  

 

However there is no way to get away from the fact that four small photosites have an inherently worse signal to noise ratio than a single photosite of the same exposed area so downsampling cannot provide the same achieved S to N ratio as a single larger photosite.  If the array combining occurred at the sensor level prior to amplification, somewhat better performance could be achieved but even the basic circuitry just to combine multiple photosites into a single output is also going to contribute some noise.  A higher ISO setting is achieved via increasing LNA gain and with that you are amplifying both the photosite generated noise and signal and also adding additional noise from the LNA which is higher at low input levels with the required higher amplification. 

 

Also, a small but significant issue is the boundary area between photosites is tiny but still existent so as you increase the number of discrete photosites you also increase this "wasted" space so even if 4 discrete sites could be read as one it still suffers an inherent disadvantage in exposed area although in observed performance this small difference is "in the noise" 🙂

 

I suspect life for a Canon engineer is frustrating at times just like it is for an engineer in any field who is trying to combine best practices in engineering with what the customer thinks that he or she wants.  I remember years ago at the dawn of the DSLR era reading an interesting short tale from a photographer in the UK who was still using his trusty EOS 1D (4 MP) several years after consumer models with much higher MP counts were common.  He was standing next to two other photographers at a moto-cross event who were very proud of and bragging about their high MP count cameras and were making fun of his 1D.  The first rider came buy and he captured 2 beautiful images while they got garbage, someone who understands equipment finds that a 4MP output can provide a pretty pleasing image (a typical HD display with 1920x1080 pixel format is only a tad over 2 MP while a dense sensor combined with a dense operator is a recipe for photo disaster 🙂  

 

Rodger

EOS 1DX M3, 1DX M2, 1DX, 5DS R, M6 Mark II, 1D M2, EOS 650 (film), many lenses, XF400 video

View solution in original post

9 REPLIES 9

Waddizzle
Legend
Legend

Varying the size of the JPEG files will only change the resolution.  Use lower ISO settings and wide aperture lenses.  There is free software than can do a decent job of reducing noise in photos.  

--------------------------------------------------------
"The right mouse button is your friend."

wq9nsc
Authority
Authority

As Wadizzle noted, for low light shooting you will get a bigger improvement from a lens instead of a new camera body.  The 90D uses a newer generation sensor than the 80D but that is somewhat offset by the higher pixel count when it comes to noise performance.  More photosites on the same sensor area result in less area per individual photosite and with the same level of sensor technology a smaller photosite is going to receive less light and will produce a lower signal to noise ratio.  With the better technology sensor, some potential for better noise performance was given up in order to have a more dense grouping of photosites.

 

Canon's 1DX 2 offers excellent high ISO noise performance from its full frame sensor with 20 megapixels resulting in a surface area of 43 square micrometers per photosite, the 90D crams 32 megapixels into a smaller sensor resulting in a surface area of  just 10 square micrometers per photosite.  Just like a wider lens aperture in a "fast" lens can capture more light, so can a photosite with greater area providing far better noise performance.

 

With the 90D, Canon appears to provide roughly the same noise performance as the 80D as ISO gets into the 3,000 plus range.  At low ISO, the new sensor can provide more detail but at high ISO most of that advantage is lost due to the need for agressive noise reduction.  The 90D has a lot of other advantages and it could have offered a pretty large jump in moderate to high range ISO perfornamce over the 80D but much of that potential advantage was lost due to the felt need to comepete in the pixel race.

 

Regardless of which camera you use, you will have a better chance of recovering the best image under trying conditions if you save in RAW instead of JPG.

 

Packing more photosites into an APS-C sensor (or any size sensor for that matter) is primarily marketing driven because more sounds better however the resolution of a 12 megapixel (i.e. 12 million photosites on the sensor) is enough to print out as large of a photo as most consumers ever will and the more dense sensors will exceed the resolving capability of a lot of the glass typically used with lower to moderate priced camera bodies.  Some current smartphones are the height of absurdity with huge megapixel counts provided by dime store glass (or plastic).  

 

Rodger

EOS 1DX M3, 1DX M2, 1DX, 5DS R, M6 Mark II, 1D M2, EOS 650 (film), many lenses, XF400 video

Yes, I understand the logic in your post. But what I'm trying to understand is why Canon can't combine results from 4 photosites to reduce noise but still provide 8MP image. Why, when I use M or S-size JPEGs it is not done. Also on some forums I read that downsampling the image helps, but never seen actual photos to compare.


@sknaumov wrote:

Yes, I understand the logic in your post. But what I'm trying to understand is why Canon can't combine results from 4 photosites to reduce noise but still provide 8MP image. Why, when I use M or S-size JPEGs it is not done. Also on some forums I read that downsampling the image helps, but never seen actual photos to compare.


Combining data from multiple photosites is not an available feature.  

 

Besides, doing so would not reduce noise, anyway.  Each photosite would contribute equal shares of both signal and noise, which means signal to noise ratios would remain constant.  Down sampling does not reduce noise for the reasons I just described.  Down sampling is claimed to improve the detail captured in an image.  It doesn't.

 

You are arguing hypotheticals and chasing phantoms.  The gear does not work they want you think it should.

--------------------------------------------------------
"The right mouse button is your friend."

They do on some cameras. Just not DSLR's. My wife's S100 has a "candlelight mode" that does this. I am guessing is that since you combine 4 photosites anyway because of the Bayer Array, it does not buy you much.

Combining photosites to get an 8 MP sensor from a 32 MP array provides a result quite similar to downsampling but does NOT provide the same noise figure as 8 million discrete photosites covering the same size sensor.

 

Downsampling works to reduce noise because the noise generated in the photosite itself AND the low noise amplifier used to increase sensor output to a usable level has a largely random distribution.  By downsampling, you are combining the noise and signal output between multiple closely spaced photosites and in best case the largely correlated signal algebraically sums and so does the noise but because the phase relationship of the noise component is randomly distributed much of it cancels resulting in less noise in a downsampled image.  

 

However there is no way to get away from the fact that four small photosites have an inherently worse signal to noise ratio than a single photosite of the same exposed area so downsampling cannot provide the same achieved S to N ratio as a single larger photosite.  If the array combining occurred at the sensor level prior to amplification, somewhat better performance could be achieved but even the basic circuitry just to combine multiple photosites into a single output is also going to contribute some noise.  A higher ISO setting is achieved via increasing LNA gain and with that you are amplifying both the photosite generated noise and signal and also adding additional noise from the LNA which is higher at low input levels with the required higher amplification. 

 

Also, a small but significant issue is the boundary area between photosites is tiny but still existent so as you increase the number of discrete photosites you also increase this "wasted" space so even if 4 discrete sites could be read as one it still suffers an inherent disadvantage in exposed area although in observed performance this small difference is "in the noise" 🙂

 

I suspect life for a Canon engineer is frustrating at times just like it is for an engineer in any field who is trying to combine best practices in engineering with what the customer thinks that he or she wants.  I remember years ago at the dawn of the DSLR era reading an interesting short tale from a photographer in the UK who was still using his trusty EOS 1D (4 MP) several years after consumer models with much higher MP counts were common.  He was standing next to two other photographers at a moto-cross event who were very proud of and bragging about their high MP count cameras and were making fun of his 1D.  The first rider came buy and he captured 2 beautiful images while they got garbage, someone who understands equipment finds that a 4MP output can provide a pretty pleasing image (a typical HD display with 1920x1080 pixel format is only a tad over 2 MP while a dense sensor combined with a dense operator is a recipe for photo disaster 🙂  

 

Rodger

EOS 1DX M3, 1DX M2, 1DX, 5DS R, M6 Mark II, 1D M2, EOS 650 (film), many lenses, XF400 video

Then a follow-up related question: Dxomark often states that superzooms like Sigma 18-300 or Tamron 16-300 have sharpness at about 7-8 MP. Does it mean that 32MP of new 90D is completely worthless for this kind of optics?

Agree with kv and take dxomark's recommendations with a grain of salt.  A "better" lens is always better but it doesn't mean that a decent lens won't work well with high MP count cameras.

 

Modern photography has gotten so ensnared in technolgy, buzzwords,and bells and whistles that many have lost focus on the basics of good photography.  I was at a football game last year and a parent had dropped a bundle on a 1DX 2 and was shooting the entire game directly into a strong afternoon sun.  Canon 1 Series cameras coupled with their high end lens can do magic but nothing is going to produce good photos when used that way.  I tried to gently and helpfully encourage him to join me on the opposing sidelines but was informed that he just purchased the best Canon had and he could shoot from anywhere.  I was also using a 1DX 2 (with a 1DX as the second body) and told him I could also shoot from anywhere but because I wanted good quality photos I wasn't going to fight a losing battle with physics and Mother Nature 🙂

 

Rodger

EOS 1DX M3, 1DX M2, 1DX, 5DS R, M6 Mark II, 1D M2, EOS 650 (film), many lenses, XF400 video
Announcements