Which is better: low resolution mode (sRAW) or downsizing afterwards?

[sereze]

I have a Casio Ex-Z 1050 compact camera, in terms of getting less noise, I did a test and found out that setting resolution on half ( 10 MP to 5 MP ) gives better results significantly. I think that some pixels don't work in low resolution mode and they don't interfere with each other and less heat etc.. I am not sure though.
I did same test with my Canon 450D, there is a difference but not as much as Casio.

Also possible to shoot at high resolution and downsize later. I don't think downsizing reduces noise but it increases signal to noise ratio or a second thought downsizing increases wieving distance and noise is less visible.

So what do you think? Am I right? And which is better, low resolution mode or post resizing?

[Colin Southern]

I have a Casio Ex-Z 1050 compact camera, in terms of getting less noise, I did a test and found out that setting resolution on half ( 10 MP to 5 MP ) gives better results significantly. I think that some pixels don't work in low resolution mode and they don't interfere with each other and less heat etc.. I am not sure though.
I did same test with my Canon 450D, there is a difference but not as much as Casio.

Also possible to shoot at high resolution and downsize later. I don't think downsizing reduces noise but it increases signal to noise ratio or a second thought downsizing increases wieving distance and noise is less visible.

So what do you think? Am I right? And which is better, low resolution mode or post resizing?

This isn't a specialist area for me, but my standard advice is to always shoot at the highest resolution possible; it's easy to down-sample to remove excess information (and reduce noise) - impossible to go the other way.

Hopefully Sean will give you a more definitive reply

[McQ]

Yes, downsizing an image greatly reduces noise. It's the same concept as discussed in the tutorial on Noise Reduction by Image Averaging. If you reduce the total number of megapixels by 50%, such as 10MP to 5MP as mentioned, you will reduce the amplitude of noise fluctuations just as with averaging two photos -- except in your case you'll be sacrificing resolution as opposed to exposure time and/or storage space.

For those interested: the amplitude of noise drops proportional to the one over the square root of the number of photos averaged. It's the same with pixel averaging when downsizing. If the original resolution is 10MP and you downsize to 2.5MP, then the amplitude of noise will generally be cut in half. This is because the number of pixels are reduced by a factor of four (and therefore 1/[square root of 4] = 1/2)

As to whether it is better to shoot in low resolution mode or to downsize afterwards: it really comes down to the interpolation algorithm used in your camera versus the one you intend to use on your computer. Back in the day (ok...just 5-8 years ago or so), in-built digital camera downsizing often took shortcuts to save on processing power, which usually corresponded to an inferior downsizing result. This would include either bilinear interpolation...or in a worst case scenario, some cameras would not even use all the original pixels, and would instead perform something similar to a nearest neighbor interpolation (which is like just picking every other pixel and discarding the rest of the image information).

There's a lot more on image downsizing here: Image Resizing for the Web & Email
In addition to a discussion of interpolation algorithms here: Digital Image Interpolation

Something to emphasize is that even if you reduce noise by downsizing the image, you are only doing so for images which are viewed at 100% on-screen. If your image has to be printed large enough that even the full resolution 10MP original is just barely sufficient (such as from 8x10 to 11x14 inches in size), then decreasing noise by downsizing will do you no good, and can in fact make the image much worse.

One last thing: the above discussion does not apply to half size image that are created by cropping out the central portion of the frame, such as with some Nikon models. In this case, noise will be identical at 100%, and worse when printed, since the digital file needs to be enlarged more for a given print size.

[McQ]

Something I forgot to mention is that you will likely reduce the number of images you can capture in burst mode by using the higher resolution setting and downsizing later, primarily because the limiting factor is often how fast your camera can write the images to your compact flash card. You'll also save a lot of space on your camera's memory card. In this sense, if interpolation algorithms are equal, it's much better to shoot in the lower resolution mode if all you intend to use is the downsized images.

Personally, I hate to throw away image information -- even if I cannot foresee any use for a higher resolution image anytime in the near future.

[sereze]

So camera actually does the same thing, shoots in high resolution and downsizes afterwards, Not what I thought like shutting down half of the pixels at the moment it captures the picture...

[McQ]

Some earlier cameras used to ignore half the pixels, as you mention, but I think it's safe to say that many current models don't do this anymore.

Edit: After a bit more reading and some thorough testing, my initial response was incorrect. In-camera and photoshop downsizing are note necessarily the same when it comes to sRAW files. I was thinking about something I'd read a while back about in-camera low resolution JPEG files. I will put some test images online later. Stay tuned...

[Colin Southern]

Sean,

I'm guessing that essentially we're talking about "Pixel Binning" here?

I've heard a suggestion that this technique was applied to the sRAW option on the Canon 1Ds3, but Canon specific comment was "no comment", and I haven't personally done any testing.

It all sounds logical - what I can't understand though is why a "bigger deal" wouldn't have been made of it if it can be used to reduce noise and thus effectively increase available ISO modes?

[Dave Humphries]

It all sounds logical - what I can't understand though is why a "bigger deal" wouldn't have been made of it if it can be used to reduce noise and thus effectively increase available ISO modes?

Isn't that just what several newer models do these days; they specifically say they go to, say 12,800 at say, half the native h x w?

I can't remember the specific makes and models involved, but I think they are making more of a deal out of it these days.

Cheers,

[Colin Southern]

Isn't that just what several newer models do these days

I haven't heard anything personally, but to be honest, I don't usually venture too far outside my 1Ds3 comfort zone.

I did a bit of "googling" on the topic this morning - didn't really enlighten me much on pixel binning, but I did find lots of other good stuff to keep me amused

[McQ]

I've heard a suggestion that this technique was applied to the sRAW option on the Canon 1Ds3, but Canon specific comment was "no comment", and I haven't personally done any testing.

Yes, after some more reading I see that Chuck Westfall at Canon will not make any specific statements on how or whether downsizing is performed, nor will he claim that there is any noticeable improvement in noise when using sRAW1 or sRAW2 compared to the full resolution RAW. The excerpt below is from an interview he did here in December 2008:

• Is the 10MP sRAW obtained from a kind of interpolation from the full 21MP? Is it true that in this way the sRAW will have a less per pixel noise than a RAW?

Chuck Westfall at Canon: The main purpose of sRAW1 and sRAW2 reduced resolution raw capture settings on the EOS 5D Mark II is to retain the image editing benefits of RAW data in terms of shadow/highlight control and white balance, etc., at a variety of resolutions and file sizes to fit the needs of professional and advanced amateur photographers. Canon is making no claims about improvements in noise levels in sRAW1 or sRAW2 compared to RAW, and the company has not published the methods it uses to produce lower resolutions. According to my tests in DPP, there is no noticeable difference in noise levels for any of these data recording modes. Since there is no significant difference in terms of noise, the main advantage of sRAW1 and sRAW2 is more images per CF card for shooting conditions where the reduced resolution is adequate for the job at hand. Nominal file sizes for RAW, sRAW1 and sRAW2 on the EOS 5D Mark II are 25.8MB, 14MB and 10MB respectively.

[McQ]

Edit: After a bit more reading and some thorough testing, my initial response was incorrect. I will put some test images online later. Stay tuned.

Ok, I've done the tests, but I did not initially believe the results. Here's samples of three images. You be the judge:

From left to right, all at 100%: (1) RAW, (2) RAW downsized 4X, (3) sRAW2

Images from a Canon EOS 5D Mark II at ISO1600 (high ISO noise reduction disabled)

My impressions are that:

• The sRAW2 image has the lowest amplitude of noise (!!!).
• The coarseness of the noise in the full resolution RAW and sRAW2 images appear very similar, although the sRAW2 image clearly has less noise both in amplitude and in coarseness.
• The RAW image which was downsized 4X certainly has the finest grain noise of all three samples, but the amplitude of this noise is slightly higher than the sRAW2 image.

Based on the above tests, I would have to conclude that Canon is not performing a simple digital interpolation when it downsizes to produce an sRAW1 or sRAW2 image. It could be that downsizing is performed in analog before the signals are converted into digital values (using hardware as opposed to software).

[_GUI_]

sRAW doesn't seem to be a simple thing. In this link in the LL it was discussed:

"SRaw is a demosaiced form. It does not need any more demosaicing, but the color needs to be white balanced and transformed from twe camera's color space in whatever.

The data is stored as a three component JPEG image; two components are subsampled. This means, that there is a green components for each pixel, but there is only one red and one blue component for each pair of pixels (in a row).

As every pixel pair represents eight raw pixels (numerically only), i.e. two red, two blue and four green pixels, every color component value in the sRaw fprmat represents two source raw pixels. The values are kept in 15-bit form; I *guess* that the purpose of this form is to store a precise average of the incorporated pixel values (the added precision covers the extra bit resulting from dividing the sum by two).

In other words the sRaw format contains *close to half* of the original values. I have not done anything with it yet, but I do expect a better quality than downresing in PP could deliver."


I want to remark the last conclusion of Gabor Schorr. Since sRAW works from RAW data, it can be optimised to obtain a better quality data summarization than achived in PP from a full size demosaiced image. In other words: it wouldn't be strange that the result of an sRAW could be better than downsizing a demosaiced RAW to half the size. In the demosaiced RAW noise will have spreaded around each pixel according the demosaicing algorithm, and also sharpness could suffer when compared to the sRAW since two interpolations took place: first demosaicing, and then re-scaling.

About the conclusions of comparing noise from RAW and sRAW files straight from a RAW developer, I would be cautious. A discussion about this has been split off here: Does digital image noise have grain like with film?

BR

[McQ]

Thanks for the link to the earlier discussion. It provides a lot of very helpful technical information. I think the answer as to whether we should use the sRAW or downsize later is *much* less straightforward than originally thought

Despite everyone's initial reactions, there might actually be situations where the sRAW is preferable (for reasons other than space and speed). That is, at least until someone comes out with software that performs the RAW to sRAW type conversion on your computer (such as an updated version of Canon's digital photo professional). I wonder if this all also applies similarly to low resolution JPEG images as well....

[kevinf]

There have been threads on the POTN forums about late model Canon cameras performing pixel-binning with sRAW. Numerous test have borne out much better noise performance using this mode, ISO 3200 becomes quite acceptable with the 50D with sRAW and you'll still have a usable file size. For the 40D, the pixel count becomes quite low with sRAW, so it'd be hard to justify for much other than web use and 4x6 prints. The 5DII and 1d(s)mkIII also give usable pixel counts with sRAW.

If noise performance is more important than pixel count, use sRAW on the later model Canons. This mode may also be useful for astrophotography and layer stacking to achieve very good noise levels.

[Colin Southern]

There have been threads on the POTN forums about late model Canon cameras performing pixel-binning with sRAW.

Hi Kevin,

Do you recall which models were allegedly capable of pixel binning?

[_GUI_]

That is, at least until someone comes out with software that performs the RAW to sRAW type conversion on your computer

That is the point, and this opens a door to think about new improved ways to develop a RAW file. If the final desired size is even smaller than half the primary RAW, some resizing strategies can be applied first to the undemosaiced Bayer data, to perform the demosaicing last over a mini-image with already the output size (e.g. for Web applications).

If sRAW can be better than full size RAW in some cases, there is no reason not to think new RAW resizing procedures could help to obtain better small images in terms of noise and sharpness.

Anyway I wouldn't expect any miracles here.

BR

[kevinf]

50d, 5Dmk2, 1Dsmk3 I'm pretty sure all bin. The 40d might as well. I'm not sure about the rebel line.

And as far as I'm aware, pixel binning needs done at the hardware level during the sensor reading, once you've gotten the RAW data, that's too late.

[Colin Southern]

50d, 5Dmk2, 1DsIIImk3 I'm pretty sure all bin. The 40d might as well. I'm not sure about the rebel line.

And as far as I'm aware, pixel binning needs done at the hardware level during the sensor reading, once you've gotten the RAW data, that's too late.

So in theory, the noise visable in a sRAW shot should be comparable to the noise visable in 4 averaged shots taken at full resolution?

[kevinf]

If I recall the results right, you get about 1 stop better noise performance. So if you are in a dimly lit area and hitting the limit of your gear, enabling sRAW could get you into the safe zone.

[kevinf]

Samples taken with my 40D:

Camera was set to manual mode on a tripod. Two shots taken, one RAW, one sRAW. Brought both shots into DPP and applied the same white balance adjustment and saved out a max quality jpeg. In camera sharpening was set to +3.



The sRAW file has been upsampled 200% using standard bicubic to match the pixel count of the RAW file.

RAW file 100% (8.22MB)
sRAW file 200% (5.83MB)


The RAW file has been downsampled 50% using standard bicubic to match the pixel count of the sRAW file.

RAW file 50% (1.97MB)
sRAW file 100% (2.49MB)



edit: here is someone else's test - http://jasonography.blogspot.com/200...-test-40d.html