Lowering Resolution on Camera - Effect on noise?

[ashleysmithd]

Hi,

I have a D40, and I wondered if lowering the resolution on camera just took the picture using the entire sensor and then down rezzed it or if it used less pixels, therefore having an effect on the level of noise?

I know you can't lower the res in raw, but it might be handy lowering the res in jpeg for when it comes to work involving high iso.

I did run a basic test, the results looked very similar, indicating the camera just down rezzes the image. Although, it was hardly a strict accurate test lol

If anyone has any insight on the matter that would be great
Thanks.

 

[JerryPH]

Reducing the MP size at the sensor reduces noise... reducing resolution size makes no difference as it is still the same sensor.

Resizing the picture size will reduce the visible noise... its still there, just smaller/harder to see.

 

[ashleysmithd]

So what do SLR's normally do? As I say from the test it looks asthough it's just taking the image at full res and downsizing, having no impact on the noise levels. However, it was a slightly crude test.. if that...

 

[JerryPH]

A dSLR will have no change in noise levels at RAW or high/med/low res levels. The noise is there all the same. It is becuase the side where the noise is created at, never changes (the sensor). The difference in resolutions is done in camera, after the noise is placed on the image thats why there is always the same amount of noise there irrespective of file format or chosen resolution.

 

[astrostu]

It really depends on how the image reduction is done, but ideally by decreasing the resolution, you decrease the relative noise. Let's say you have 4 pixels, and the light recorded on them has a value of 100 photons. The noise on each of those pixels is ±10 photons (Poisson statistics, noise is SQRT(counts)). That's a 10% noise level. When you add them together (bin by 2, so reduce size of image by 50%), then you add the counts together (400) and divide by 4 (so you're back at 100). The noise, however, gets added in quadrature, such that the final noise level will be SQRT(10^2+10^2+10^2+10^2)/4 = SQRT(400)/4 = 20/4 = ±5. So, your final uncertainty, after reducing the size of the image to 50%, will be at the 5% level instead of the 10% level.

This is why astronomers almost always down-sample their images, binning in 2x2 or 4x4, to decrease the noise.

 

[Dwig]

...
This is why astronomers almost always down-sample their images, binning in 2x2 or 4x4, to decrease the noise.

... or they stack a series of separate images.

In simple terms, since noise is random combining photosides, either neighboring photosites when downsampling or matching photosites when stacking multiple images, produces a pixel that is the average of the combined photosites. Neighboring pixels are more consistant and thus there is less noise.

This is now a common trick in a number of P&S cameras. You'll find that many have higher ISOs available than previous generations, but on close examination of the specs you find that those higher ISOs are onlly available at lower resolutions. The reason is that downsampling reduces noise and only this reduced noise makes the higher ISOs usable.

And to answer one of the OP's questions that's been ignored, when you select a lower resolution on a digital camera the image is still shot at full resolution. The selection of lower resolution merely causes the software in the camera to downsample the image before saving as a JPEG or TIFF (RAW is always saved at full resolution).

 

[Garbz]

Reducing the MP size at the sensor reduces noise... reducing resolution size makes no difference as it is still the same sensor.

Resizing the picture size will reduce the visible noise... its still there, just smaller/harder to see.

Sorta wrong on both accounts. It depends on HOW the image is reduced. If the camera reads out every second line to quarter the image resolution then you have no noise reduction. But if the entire sensor is read and the image downsampled then you do get a noise reduction.

Resizing the picture does not make the noise harder to see, it physically reduces noise. Noise is a Gaussian distribution centred around zero. That means that two adjacent pixels have would have different noise levels, one higher than the other. When you reduce the image size the resampling algorithm would not just drop one pixel over the other. It would examine them both and then determine the appropriate value based on whatever windowing function is applied (i.e. Bicubic smoother). The end result is the noise is averaged (reduced).

/Edit: And now that I read the whole thread astrostu got in first.

 

[JerryPH]

Reducing the MP size at the sensor reduces noise... reducing resolution size makes no difference as it is still the same sensor.

Resizing the picture size will reduce the visible noise... its still there, just smaller/harder to see.

Sorta wrong on both accounts. It depends on HOW the image is reduced. If the camera reads out every second line to quarter the image resolution then you have no noise reduction. But if the entire sensor is read and the image downsampled then you do get a noise reduction.

Resizing the picture does not make the noise harder to see, it physically reduces noise. Noise is a Gaussian distribution centred around zero. That means that two adjacent pixels have would have different noise levels, one higher than the other. When you reduce the image size the resampling algorithm would not just drop one pixel over the other. It would examine them both and then determine the appropriate value based on whatever windowing function is applied (i.e. Bicubic smoother). The end result is the noise is averaged (reduced).

/Edit: And now that I read the whole thread astrostu got in first.

If I take a shot on my D700 at ISO 128000 and JPG fine, it is VERY noisy. I will take a look at the full sized JPG at the areas which have the most noise then reduce that picture by 50% in post. On screen, the shot then "looks" cleaner, but the same noise is in the same places, but is less visually visible when I bring it up to 100% view on screen. The size of the pixels may have been reduced and that will hide it some, but that doesn't mean it removed it. The same shot on the camera using FINE or SMALL jpg files resulted in near visually identical amounts of noise when comparing them side by side at magnification levels so that the pictures appeared equally sized (ie: 25% on the large file and at 60% on the smaller file). This to me tells me that changing resolution in camera does nothing to improve on reducing noise.

I've played a lot with this on my D200 too before I purchased the D700. I would often use little to no NR, but resize the file in post process and it would appear better, but when taking the same shot at high ISO at any of the different JPG size settings, I always saw the same amounts of noise there. I'll try to bring up a few examples of what I did this weekend if I can find the time (I have a wedding on Saturday and an engagement session and a "Trash The Dress" on Sunday, so we'll see).

Granted, I do not know how all cameras rasterize the image at the sensor, but I do know that for at least the D200 and D700 changing camera resolution doesn't do anything to actually give me a cleaner image... short of reducing ISO and nailing the exposure.

 

[astrostu]

Thanks, Garbz, though you and I were talking about slightly different sources of noise (me in counting statistics, you in sensor noise), it's the same situation where, if the camera does the math right and actually takes advantage of the way you should bin data, then shrinking the image will result in a reduction of noise.

Jerry, I would suggest that if you really want to try to measure this in a qualitative way instead of just by eye (and you really have to do it in a qualitative way, doing it by eye will not give you an objective answer), then you take a shot of a neutral 50% grey object (128 brightness in 8-bit) and you actually look at the numerical value of the pixels under various settings (small/large JPG, high/low ISO). If the camera does the image the way it should to take advantage of really basic statistics, then there WILL be a reduction in the noise when you bin it.

Alternatively, the camera may not actually bin the image when you shoot at lower than native resolution. It may just sample every-other pixel. Or every-third. Then that would NOT decrease the noise.

 

[JerryPH]

astrostu, though I would like to measure it out in your way, the only way available to me are my own eyes... and they tell me that there is no visible noise reduction between shots that are taken on the same camera at different resolutions (RAW > JPG vs JPG fine vs jpg small).

It is possible that the smaller sized JPG is generated in the manner you suggest, I just have no way of finding that out.

I would also tend to think that if this was true, everyone who's primary concern was low noise in darker environments would not shoot raw, but would shoot small JPG, which to my limited knowledge... no one does.

 

[itznfb]

I would also tend to think that if this was true, everyone who's primary concern was low noise in darker environments would not shoot raw, but would shoot small JPG, which to my limited knowledge... no one does.

actually quite a few people do. i've met a lot of hobbyist zoo photographers recently and almost all of the shot jpg medium on 10+ mp cameras. i'm the only one so far that shoots raw. however, we never got into why they used those settings, so it may not be for that reason.

 

[JerryPH]

actually quite a few people do. i've met a lot of hobbyist zoo photographers recently and almost all of the shot jpg medium on 10+ mp cameras. i'm the only one so far that shoots raw. however, we never got into why they used those settings, so it may not be for that reason.

I bet it was simple lack of knowledge. Best quality shots are done in RAW, not small JPG .

I do a LOT of shooting in low light scenarios almost EVERY Saturday for the last several months (doing weddings). I just could not see myself switching to JPG small at the reception just to get cleaner shots... lol. I regularly shoot at ISO 3200 and the odd ISO 6400.

 

[astrostu]

Jerry - itznfb got to it before me a bit, but your comment makes three fundamental assumptions that I think are not true: First, that photographers know about counting statistics and sources of noise in digital electronics, second that they don't do this, and third that they would rather the camera do the processing than them do it.

To the first point, I really don't think many photographers - especially the "old guard" who shot film for decades and are moving to digital - know about this. I do because I've taken years of astronomy classes.

For the second, as itznfb said, people do.

Third, I prefer the camera do as little processing as possible since I don't know exactly what it's doing. I prefer to shoot in RAW at native resolution and then do any down-sampling myself. That way I know what I'm doing and what happened to the picture during processing and I can make sure that it's done right.

 

[Garbz]

My example was an over simplification Jerry. It's all statistical in nature and also assumes that you're using a perfectly averaging resize. I think Bilinear being a very smooth and a not often used for photography sampler would give the best result noise wise, but also the softest image.

When resizing the statistics only tell that the average noise will be reduced. Obiously it's perfectly reasonable that two adjacent hotter pixels exist and will averaged to have the same hot nature. The statistic does not say alternating pixels are hot and cold in a perfect pattern (in which case resizing would effectively eliminate ALL noise).

Also on the topic of resampling, many filters use a post sharpening effect or have such an effect as a result of it's function. Bicubic Shaper for example which is the photoshop nominated "Best for reduction" has a natural overshoot which causes haloing of adjacent pixels increasing the sharpness. In other words Bicubic Shaper would clobber the noise and increase it by sharpening it at the same time.


But this is just image processing theory. How the camera does it is anyones guess. I've never actually tried to shoot at S or M with my D200.

 

[JerryPH]

Just to tell you that I too never even shot JPG on the D200, much less S JPG... it actually took me a few seconds to even find it in my menuing system of the D200... lol

I think that one can talk technical theory until they are blue in the face (and please that is NOT any kind of dismissal to the good info offered here by astrostu and garbz), but in the end, when you place 2 pictures side by side and do a little pixel peeping and see which has less noise, that is what really matters, at least to the average layman like myself.


D200, SOOC, no noise reduction anywhere... the ONLY changes to the pic was the reduction of the RAW file image (after it was converted to a JPG, of course), so that it could be downsized to approximately the same size as the small JPG:

The test baseline, RAW ISO 100:

The RAW file at ISO 3200:

The SMALL JPG at ISO 3200:

Both ISO 3200 shots are pretty bad (lol), but from what I can tell, the RAW file had more artifacts, but appears even a little cleaner than the JPG file?? I expected them to be the same... not see a slight improvement over the JPG. :meh:

Please forgive the minor changes in exposure, I had the camera in aperture priority. I do think this skews the tests a little in terms of noise reduction amounts, but not all that much to invalidate the overall outcome of this very quick and unscientific test.

I invite others to try out the same test and see if the results are the same or different for you.