Noise in photos and its causes

R

Rich

Messages
435
Edit My Images
No
Hi All,

I have noticed that more or less everytime i go out shooting i come back with a shed load of shots that are unusable due to noise, especially in dark areas, but often in light areas too.

For example, today i was in the Lakes, and ive come back with 80% noisy crap. The light was good, and the highest ISO i used was 400, but varied from 100-400.

What other considerations do i need to make to reduce noise in my shots?


Thanks! :)
 
Well you've described the causes Rich - high ISO and low light are the main causes of noise in your images. Noise will also be seen most in the darker part of your images.

I have to say I often use ISO 400 and usually don't find the noise that noticeable, in fact it can often resemble grain in film with modern digital cameras which isn't always unacceptable or unsightly.

Do you try running these shots through noise reduction software? :)
 
thanks for the info. Yes i probably will try noise reduction software at some point, but the amount of noise im seeing is quite bad and id like to get things sorted before post process. I think i need to work on fine tuning exposure as well as this could be ruining things for me.

Cheers!
 
I suppose it's quite possible too that polarisiers and other filters needing an exposure compensation increase can contribute to noise, by cutting down light hitting the sensor and lengthening the exposure time.
 
Rich said:
thanks for the info. Yes i probably will try noise reduction software at some point, but the amount of noise im seeing is quite bad and id like to get things sorted before post process. I think i need to work on fine tuning exposure as well as this could be ruining things for me.

Cheers!
Click to expand...
Well under exposure will lead to increased noise Rich. Can you post an example of one these shots?
 
Steve was telling me the other week about a forum he was reading which went into great detail about noise.
Apparently it got very very technical and geeky and went on to discuss frequencies etc etc.

Basically what they had found that (I forgot the details), but the jist of it was that the noise is there throughout the ISO range, irrespective of what ISO you're using, there's the same amount of noise in the shot at ISO100 as there is at ISO3200. It's when it becomes evident or something.
The upshot was, that if you trust your histogram, the trick to using high ISO's was to overexpose your image, obviously not blowing out the highlights (this is where trusting your histogram comes into play).
Then during the raw conversion process, you can drag the exposure compensation back down to return the shot to normal levels. Something like that.

Like I say though, it was mentioned over a 5 minute conversation, but Steve said either he'd taken or seen some pretty much noiseless shots at ISO3200.

Disclaimer : All of the above is from memory so could very well be factually incorrect, but the jist of it is true. At high ISO's, overexpose the shot. Also, I've slept since that conversation too, which is my brains reset button :D
 
I was using a polariser today! Probably did have a large affect on it given the varying light levels (mountains = sunny in spots and masses of shadow in others).

Ill try and post an example when i go throught this lot :)

Ta.
 
Marcel has basically got the gist of it and its all new to me as well...what was being said in a very geeky manner and far beyond my technical knowledge is that if you shoot in raw, put all your in camera parameters (saturation, contrast etc) to the absolute minimum (minus settings) as they are only valid for JPEG photos and after the camera has “processed” the file, then over expose to the right but not past a recoverable amount. This increases the gain of the sensor and records as much detail from the shadow areas and from the blue channel (which actually is about 10 times less efficient than the others) giving you masses of detail to work with. It also virtually eliminates all the visible noise.

I was sceptical but when I was in Belfast Zoo with Stewart I altered my settings and gave it ago. The tricky part is getting the amount of over exposure just right as if you go too far to the right you will lose the highlight detail and blow out the picture beyond a recoverable amount. This happened to me on a couple of my shots but for the most time I managed to get it about right. If you are going to attempt this method of reducing the noise I would highly recommend practicing a lot first and on several different lighting situations, I did it on animals at the zoo as it didn’t really matter if I ruined all the shots as I had already recently managed to get some good animal photos from our Chester zoo trip.

IMG_4391_01.jpg


Canon 20D, ISO 800, 1/1000sec, F5.6

IMG_4420_01.jpg


Canon 20D, ISO 1600, 1/640sec, F5.6

and to show the blue channel noise ;)

IMG_4465_01.jpg


Canon 20D, ISO 800, 1/160sec, F5.6

All the above have just been processed in raw shooter and then resized, they have not been run through any noise reduction files.

Now if that doesn't start a discussion...
 
That's very interesting Steve - I'll have to give that a whizz! :) So - are you saying the bulk of the noise is carried by the blue channel? (being less sensitive)
 
I am not saying that, tbh most of what I read was very technical and way beyound me but the bods who were discussing it were. What I did glean from the conversation is that apparently sensor noise is fairly constant and its only the lack of detail at higher iso's that allows it to become visible.
 
Got a link to that discussion? Being a computer bod (and there are others on here I suspect) we might be able to translate it!

BTW Steve, what does putting all the settings down to zero do if they only apply to Jpegs? Or did I miss something there?
 
no link at the moment..running on the lappy as the main rig is still packed from the trip over here. i will dig it out once I get sorted though.

Putting the settings down to minus figures best represent an unprocessed JPEG ie the closest you can get to a raw histogram which is what you need to work from to get the exposure as close as possible. The Histogram shown on the camera is extracted from your in camera settings for JPEG shooting even if you are shooting in RAW. This is done automattically by the camera so its the next best solution to giving the most accurate histogram for raw that we can get. Normally the difference wouldn't matter that much as we have lattitude when shooting raw but since we are aiming to use all that headroom, anything we can claw back is a benefit.
 
Just a quick question guys,

At what stage of processing is it best to run a image through noise reduction software, and is there any adverse affect by doing it before you save your image for uploading to the site.

..........and can anyone point me in the direction of a link a simple to use "noise reduction software" site please?

Thanks

Mike
 
My workflow consists of the following :

Rawshooter
PSCS2
Minor absolutely necessary edits (rotate, etc)
Noiseware
Then I work on the image more if needed, although alot of my adjustments tend to be done in RSE if I can help it.
 
SammyC said:
Got a link to that discussion? Being a computer bod (and there are others on here I suspect) we might be able to translate it!
Click to expand...

Took me a while but as promised…

Most people assume that because the camera takes noisier pictures at higher ISOs in automatic mode with constant exposure compensation (or none), that noise is directly related to ISO, but this is not really true when you look at things from other perspectives. The signal-to-noise ratio in the sensor is the same, no matter what ISO setting you use, if you use the same f-stop and shutter speed on the same subject in the same light. Up until the point that the camera starts converting the analog voltages in the sensor to RAW numbers, ISO only affects one thing - metering. So, if you are managing the absolute exposure yourself, the ISO setting has no effect on the analog capture. Once the camera starts reading the sensor, then ISO makes a tremendous difference, because it causes a different range of sensor voltages to map to the numbers 0 through 4095. Given a specific absolute analog exposure on the sensor, you get the cleanest RAW data from the *highest* ISO that doesn't clip your highlights. This is not very useful information for high dynamic range images, like dark grey things in the shade and white things in the sun in the same image, but in medium- and low-contrast scenes, you can expose way to the right without blowing RAW highlights. JPEG settings don't affect RAW capture, but if you set the contrast to -2, your histogram become a bit more accurate for RAW data. Assuming that the WB setting is close to reality, white highlights generally clip in the 20D histogram just before the point where they clip in the RAW data.
Click to expand...

Now if you want to read through all 4 pages of posts which go into far more detail than that small extract, then the whole thread can be found here. Be warned it gets quite heavy and is difficult to follow at times.
 
OK, I've read through the four pages of the thread, applied a large (LARGE) cup of tea to my brain and I think I've worked out what the heck they were on about.

If anyone is interested then I'll attempt to write it up and explain it (as I understand it) in non-technospeak.

:)
 
A quick experiment I tried, camera on full manual only change is ISO and exposure is corrected in RSE:

ISO100
ISO100.jpg

ISO1600.jpg

ISO1600

Clearly there is less noise here at 1600 than 100.... verrrry interesting. :)
 
The ISO 100 image is darker though Sammy, so will tend to show more noise.
 
Where's that chin scratching smiley?

What settings are those shots?
 
CT said:
The ISO 100 image is darker though Sammy, so will tend to show more noise.
Click to expand...

Only marginally. The originals out of the camera are obviously very different.

Marcel said:
What settings are those shots?
Click to expand...

Both were taken at f/3.5 and shutter 1/50, i.e. constant exposure.


The discussion that Steve found states that the ISO setting has nothing to do with the sensor, it captures the same amount of light regardless of ISO. The ISO comes into effect during the analogue to digital conversion phase.

If people are interested I can take the time to write it up but it's quite complicated to explain so I'll only do it if people want to know, it's quite geeky and could be a bit :yawn:

:)
 
If I understand this correctly the 2 images both had identical 'quantities' of light hit the sensor.

One shot had the picture brightness amplified in software (iso 100)

One shot had the picture brightness amplified in the cameras a/d convertor (iso 1600)

So the camera is better at amplification of a weak image than raw software?

Or is there more...
 
That's kinda it Rob.

To be honest it took me a while to fully get this so I'm concerned about giving a little information and misleading people than taking the time to explain it properly and giving people the full information for them to take on board.

In my 'tests' it took me a while to get an image that didn't look better at ISO100 but this was because I didn't fully get the implications of the problem.

I'll try to explain this through the medium of pictures (if you want the medium of dance you'll have to pop round) in a bit. But feel free to ask questions and I'll do my best to explain them if I understand it!

:)
 
The concept of "shooting to the right" is one that can (and many say should) be adopted in all circumstances as it maximises the data you get out of the sensor.

The whole deal gets very techy and to be honest I find it hard to retain any more of this stuff than I need to produce the goods but I recall it something like this......

The way a digi sensor records what it sees produces far more detail at the highlight end than the shadows. In fact, half of all the data captured is in the first stop of the histogram. Then half of what's left is in the next stop and so down through the (typically) 5 stops.

So if you say a sensor is capable of producing 5000 values between absolute white and black, then you have 2500 in the first stop, 1250 in the next and just 156 for the deep shadow detail. This means that if you're shooting a scene that requres less than the full dynamic range of the sensor you can increase the detail you get by tucking the highlights up the histogram as far as you can. Then when you reset the values in processing to their correct places, you have a better shot than a correctly exposed one.

It makes sense that this continues into noise reduction but the problem is one of, works well in the theory but not always in the field. Sure if you are able to shot to the right then you'll see the benifits but often we are ramping up the ISO because we don't have the luxury of choice.

I mean how many times have you pushed the ISO as far as you can and still been left wanting more light? I know that for sure I'll be keeping in mind that a shot overexposed at 1600 ISO can be better than the correct exposure at 400 ISO but as is always the case with photography.........

... there's never one answer. :LOL:
 
Sammy I think I get the jist of it, but would love to hear a laymans explanation (or as layman as you can give of course).

:)

This surely is fascinating stuff.
 
'Kin Hell Daz gimme a break - I'm only a Walsall chap! :ponders:

:LOL: I'll try to get me 'ead round that a bit later on.
 
Ok, this is my basic understanding of the process:


Light arrives at the sensor cell, the amount is determined by shutter and aperture.

Each cell holds a charge from the arriving light and that is then converted to a voltage.

The analogue half of the process can be assumed to have infinite resolution in the values it can represent. For example if the sensor cell can output a range of voltages in the range of 0 to +5v then there are an infinite set of values in between the two limits.

Up until this point the ISO has NO effect on the process.

The ISO setting is now applied before we perform the AD conversion. It is effectively a gain setting applied to the voltage amplifier before being quantised into the digital value. What this means in English is that the ISO setting is a multiplier, ISO100 could be 1x, ISO200 2x, ISO1600 16x etc.
Note: These voltage ranges and ISO scaling factors given here are just assumptions made on my part to allow the examples to be given. The actual ranges and gains used will be particular to a specific camera and sensor.

So a value of 1.1V from the sensor with an ISO400 might end up be passed to the AD conversion unit as 4.4V (1.1V x 4). If ISO800 (8x) were to be used then 1.1V scaled by 8 would equal 8.8V, but our range has a maximum of 5V so it would be clipped to 5V effectively giving a blown highlight in the picture.

This voltage is then passed through an Analogue to Digital conversion unit to arrive at a 12bit value (range 0 – 4095). 0V = 0 and 5V = 4095.

All the 12bit values are collected together to form the RAW file. JPEG images has nothing to do with this process except that it is an 8bit scale rather than the 12bit scale of the RAW.


So an example:

For a low light picture, at ISO100 (no scaling in our camera) the minimum analogue value read could be zero and the maximum might be 5/16th or 0.3125V.

When this is converted to a digital 12bit value (range 0 to 4065) this gives us a dynamic range of 0 to 255, 0V is 0 and 255 is 5/16th of 4095. So, any of the pixels of our analogue image are going to be quantised onto a range of 0 to 255.

In our example two adjacent cells hold the voltages of 0.3125V and 0.3114V. These get converted to 255.9375 and 255.0336, but because this is digital then both become 255.

So the subtle differences between these adjacent pixels would be lost by the rather crude range being used at ISO100.


If we switch to ISO1600 (x16 scaling in our camera) for the same image, the minimum analogue value after analogue amplification is still zero (16 x 0 = 0) but the maximum is now 5V (16 x 5/16th = 5).

When this is converted to our 12bit value the full range of analogue values are used and therefore the full range of digital values are used, 0V is 0 and 5V is 4095.

In our example the two adjacent cells holding voltages of 0.3125V and 0.3114V are first amplified by the ISO1600 gain value to 5V and 4.9824V respectively. These are then converted to 4095 and 4080 on the digital scale.

So the subtle difference between these two cells has survived the conversion to digital and we should have a smoother image at the end.
 
noise is definately a product of amplification. basic electronics I think. Hence why compacts have more noise than a Dslr. The photosites are smaller and therefore less sensitive to light, so have to be amplified more and therefore more noise.

Heat also causes noise. I.e. as the photosites heat up from absorbtion of light they will gradually produce more noise. hence why compacts use noise reduction for longer exposures. the sites heat up overtime.

that's about as much as I know to throw into the debate. I don't think it'll help much though!
 
Told you it got quite geeky and over complicated.

The question is ...

is it better to shoot at a higher ISO and get a well expossed picture (possibly exposed to the right) or shoot at a lower ISO with a slower shutter speed and under expose, later to recover the pic in software or during the raw processing...which would typically produce a photo with less noise?
 
higher iso. Longer exposure=higher photosite temp=more noise.

but that may depend on the original size of the photosites, quality and size of sensor, the way the camera amplifies the raw signal from the sensor etc.

Suck it and see I say.

My fz30's noisy anyway. alledgedly. Will get round to posting some shots when I actually get outside and it's not raining. :(
 
What you are trying to do is not expose left or right but just make use of the full range of the RAW colour space. So, you use the ISO gain factor to try to widen your histogram to fill the full range, and expose correctly. In some circumstances you have a very high contrast image that has a wide range automatically, so no scaling of the range is possible so we use a low ISO. If you have a low contrast image then a high ISO can scale the values to fill the full range.

You should use the highest ISO you can without blowing any highlights. On a bright sunny day that might mean that ISO100 is your only option.

So, as I work it out Steve, it's preferable to get as wide a range of values as possible. In order of effectiveness (i.e. less noise):

1) More light, either by it being there or a long exposure
2) High ISO, so doing the light value amplification at the analogue stage in the camera
3) Under exposing and boosting at the digital stage in software.

This is because as you move down the path from light source to RAW file you are losing information, digitisation is inherently lossy.

But this is all theory!
 
SammyC I understood that explanation, thanks!.
 
Don't you just love all this theory and practice stuff, you learn as much as you can, read the manuals, scan the web, talk to people then go out and practice were you find that all the theroy makes no sense ;)

Thanks Sammy for wading through it all and trying to help out us dumb people. :thumb:
 
That my thinking too Steve! I can see how this ISO thing works, I can see the implicatinos of it, but I have no fecking idea how to apply this etc. lol

If I get the urge I might try to write this all up for putting somewhere, seems a shame to lose this now that we have the beginnings of an understanding.
 
Marcel said:
SammyC I understood that explanation, thanks!.
Click to expand...

My pleasure! It's nice to be able to 'give something back' after it's just been take take take so far!
:)
 
gandhi said:
higher iso. Longer exposure=higher photosite temp=more noise.

but that may depend on the original size of the photosites, quality and size of sensor, the way the camera amplifies the raw signal from the sensor etc.
Click to expand...

That's a different type of noise than I was talking about. In fact what you are talking about is more true noise rather than the posterisation noise that I've been talking about which is more deresolution issue that just looks like noise.

Do you think this has much effect on the image then?
 
I think the test would be to take a short expsure, a very long exposure, and then another short exposure and compare the noise in the two short ones.

It may have an affect on smaller sensors, but when you get to aps-c sized and full frame sensors, i think the difference would be negligible.

My FZ30 is noisy relative a dslr. but that's cos it has 8 Mp on a 1/1.8" sensor. back to what I was saying about the size of the sites and noise, as the pixel density is lower on larger sensors, there is also less leakage of current from one site to another. again, less noise.

if you try and pull shadow detail you will always end up with more noise as less light hitting the sensor means a lower voltage and therefore a lower s/n ratio.

I think a lot of the stuff abut heat etc is theoretical and has little real world influence.
 
I understand what you are saying gandhi but this must also have some effect on the results gained from DSLR's with or without full frame sensors. Higher ISO has traditionally produced higher noise, what is being said (I think) is that if you get the exposure correct even at higher ISO's there is no greater noise. Its just proving it and wading through all the theories to get the facts with real world examples and then to be able to use that in our daily photography.
 
Depends which 'noise' you are talking about there Steve, Gandhi's heat noise or the rounding error noise that you originally introduced and that I have continued with.

For the second case, the rule of thumb is that there is less 'noise' in higher ISOs (up until clipping occurs) for a given exposure (constant shutter and aperture). I think you will always get a better result if you have control over the exposure and use a lower ISO though.
 
You must log in or register to reply here.
Back
Top