Canon 50D noise worse than 400D

<p>By virtue of its larger pixel size, the 400D blurs the noise more than the 50D. You can achieve the same effect on a 50D image by downsizing it to the size of a 400D image. So in low light conditions, there may be no advantage of one or the other (aside from the higher ISO speeds of the 50D). But this is only one usage scenario.<br>

The real advantage of the 50D's higher pixel density is in adequate light with lower ISO speeds (perhaps ISO 400 or less), with a good, sharp lens.</p>

I havent use any of these cameras. But, my opinion is, if u view at pixel level 400D is less noisy than 50D. But if u compare same sized outputs, 50D is less noisy than 400D. See this page: http://www.dxomark.com/index.php/eng/Image-Quality-Database/Compare-cameras/(appareil1)/267|0/(appareil2)/184|0/(onglet)/0/(brand)/Canon/(brand2)/Canon

Go to SNR 18% tab. To compare pixel level noise, select "Screen". To compare noise in same sized outputs, select "Print".

For some of us, the best camera would consist 4 pixels only: 2 for green channel, 1 for blue and 1 for red. The pixels of this camera will be largest ever. In this way, we wil get a camera which will have highest full well capacity and lowest possible noise. But what we can do with the output of this camera is anothor question. Right?

<p>This issue has been beaten to death ever since the 50D came out. The only consensus seems to be that the per-pixel noise of the 50D is more than the 40(0)D. This is not as inevitable as some make it seem because the seamless microlenses should, -according to Canon - make the effective size of the photosites similar. Apparently, that didn't work.</p>

<p>Then a lot of people disagree on whether or not prints from high-iso shots on a 40D look better or worse than those from a 50D shot under similar circumstances. I have not seen any direct comparisons, but my take on it is that the difference can't be very large if so many disagree.</p>

<p>IMO, the real question that has to be worked out, is how many extra pixels do you need to compensate for a given amount of extra per-pixel noise. Most with an engineering background swear by the square root law (e.g. the clarkvision website quoted above, or DxOmark). Although Aresh tells us we are not to disagree with that, I will go out on a limb and say I am not an average Joe. I don't think it applies, because</p>

<ol>

<li>Demosaicing introduces correlations between the pixels which invalidates the 'uncorrelated, random noise' assumption that underlies the square root. This is probably a minor point, and it really shouldn't matter if the noise is measured in RAW files. </li>

<li>Noise is always measured as an RMS deviation. But our perception of noise is not the same as the RMS deviation. Compare a grey image with 1% white pixels to a grey image with the same RMS noise, but random around the grey value. It doesn't look like the same amount of noise. Strong outliers, like dead or hot pixels, catch our eye. So does patterned noise, e.g. the stripes you get from readout in digital cameras.</li>

</ol>

<p>I guess this is related to what DPReview were trying to get across in this article: <a href="http://blog.dpreview.com/editorial/2008/11/downsampling-to.html">http://blog.dpreview.com/editorial/2008/11/downsampling-to.html</a> . But I don't think they really went to the bottom of the issue yet.<br>

So, any scientists out there who deal with image perception? Is there such a thing as a known weighting of noise amplitudes? If not, start writing your proposals please.<br>

Allard</p>

<p>-</p>

<p>Allard,<br /><br />A few points for you:<br /><br /><em>1)</em> noise is measured from RAW data pre demosaic so it is uncorrelated (white).<br /><em>2)</em> CMOS image sensors subtract dark current on chip, so residual noise has four components:<br /><strong>a)</strong> Photon Shot noise (Ultimate QM limit of any detector without time-averaging). <br /><strong>b)</strong> capacitive cross talk between row and column pathways.<br /><strong>c)</strong> lateral optical cross talk at CFA.<br /><strong>d)</strong> Johnson’s noise from wiring and sense row/column amplifiers.<br />From the components above, SNR due to photon shot noise scales with pixel pitch (SNR=sqrt(N), where N is the # photons absorbed) or Poisson's distribution. If the other components were zero then 50D/40d SNR would be ratio of pixel pitches (5.7um/4.7um) ~ 1.2, the value that R. Clark has measured (Fig. 2) in his article, is 90/70~1.3 which is very close. This indicates that current generation of CMOS sensors are very close to shot noise limit. For the sake of argument, you are left with a factor of (1.3/1.2)^2 to cancel out with scaling which is about 17%. The 40D was 10 Mpixels, so the optimum point is about 12 Mpixel. <br />Also noise is measured via power spectral density at given wavelength(s) (since # of photons absorbed in semiconductor material per unit area is a function of wavelength per mode) for example 18% grey. Since noise is white, you can plot an intensity histogram and measure the corresponding standard deviation of the resulting Gaussian distribution, the famous "Bell curve".<br />Human perception if noise however is a subjective factor and thus, cannot be quantified, it depends to healthiness of eye and brain. Some people may find noise distracting while some may be totally blind to noise. If you are interested more in learning about sensor architecture and noise measurements and have the required back ground in EE, take a look at this excellent article that one of our labs published a few years ago.<br /><a href="http://isl.stanford.edu/~abbas/group/papers_and_pub/isscc02_tutorial.pdf">http://isl.stanford.edu/~abbas/group/papers_and_pub/isscc02_tutorial.pdf</a><br />Best,<br />AH</p>

<p>typo, above should read SNR (40D/50D)=1.3, sorry</p>

<p>The truth is that Canon went a little to far with pixel count on the 50D and c-sensor tech in spite of the fact that "gapless microlenses" were suppose to offset noise. At this time 12mps seems to be about the limit before you enter the realm of the res for noise trade. pith</p>

If we rank Canon DSLRs Signal to Noise ratoiwise, the list will look like this (from highest SNR to lowest SNR camera):

1. 5D Mk II

2. 1Ds Mk III

3. 1D Mk III

4. 50D = 500D

5. 450D = 400D = 1000D (= 40D)

<p>Both images look the same and are pretty noisy.</p><div></div>

<p>Wow. As someone who actually uses a 50D and read every review I could find before buying it, I find the level of vehemence in this argument weird.</p>

<p>First, maybe other people care about other things, but I care about the appearance of my final images, period. Pixel peeping is just a means to an end, in my view. So the argument about whether an image should be the same size in comparing two cameras with different pixel counts seems utterly pointless to me. If your goal is understanding the functioning of the sensor, the answer is no. If the goal is understanding the quality of your image, the answer is yes. If you have in mind an image of a given scene at a given size, the relevant comparison is that image at that size, produced with two different cameras. Blowing up images of different resolution to 1:1 is an informative exercise, but it is not the one that counts in the end.</p>

<p>Second, it's worth paying attention to the magnitude of differences. What matters to me is effects that are readily apparent in the final images.</p>

<p>From that perspective, my reading of the reviews was that the practical difference in noise between the 40D and 50D is pretty minor. I never used a 40D, but I have shot a fair amount under various circumstances with a 50D, and the story seems pretty clear to me. Noise at 100 and 200 is simply negligible, regardless of how I shoot. Noise at 400 is sometimes noticeable, particularly if I expose to the left, but it is rarely bothersome enough to use anything more than the weak NR in LR. 800 definitely is noisier than I would like, but still usable, although I avoid it if I can and end up using Noise Ninja more at that ISO. More often, but still all that often.</p>

<p>When I bought the 50D, I thought I would have been happier if they had done 12 MP with less noise. Now I am not so sure. The difference in noise would not be great, and I have to say that I am really enjoying the ability to crop more severely.</p>

<p>Dan</p>

<p>oops--apologies to the OP. I lost track of part of your question. When I got the 50D, I replaced an XTi, which I think is a 450D. I did not try A/B comparisons of identical images, but my consistent impression is that the noise of the 50D is less apparent than that of the XTi, for comparable images. Certainly not more apparent.</p>

<p>Aresh, I know all that stuff, that wasn't the point. BTW, isn't it so that where it matters, in the shadows at high iso, the noise is limited by readout, not shot noise? Anyway, the thing I disagree with is</p>

<blockquote>

<p>Human perception of noise however is a subjective factor and thus, cannot be quantified</p>

</blockquote>

<p>There's variation between individuals, but that doesn't mean quantification is impossible.</p>

<p>Somewhere amongst all these assertioins lies the truth. It should be based in science. First question is there a generally agreed upon standard for measuring noise? Then is there a process for making valid measurements against that standard? Then is there someone who can construct scientifically valid tests that can be replicated and withstand review by scientists who have published in the field? I am not sure what I read above meets that kind of rigidity or disciplined research. </p>

<p >Allard first my name is not Aresh. Secondly It depends to how dark the shadows are, electronic noise is less than 5 electrons so as long as you absorb more than 25 photons per pixel you are limited by shot noise regardless of ISO, a typical warm shadow emits way more than that and the cut off filter on sensor still has finite transmission in shorter IR. If you know the stuff why are you asking any way? </p>

<p >This is yet another wasteful thread, I am done. I leave it to the Pulitzer-prize-winning photographers and Nobel-prize-winning physicist of this website to decide how to measure noise and such. </p>

 

<p>Some in the industry think 13mp is about optimal for a DX sensor using current photo cell technology. (Look how well the D700/D3 does with 12mp!) Maybe manufacturers are able now to make sensors with fewer but much better cells thus reducing the noise. This may well be why Canon went down to a less dense sensor with the G11. They face a marketing fight, though.<br>

Finally, we might be seeing an end to the mp count as the indicator of image quality. The 40D was always going to be a hard act to follow. Once in a while every manufacturer makes a really good model which seems to get the mixture of features, quality and price just right. For Canon, that was indeed the 40D. This body was almost always the back up camera of choice with Canon pro's after the 1D. In fact, I know some who have sold their 1d's and use two or three 40d's. That says a lot.<br>

But I am sure, in a final bigger-than-A4 print, the 50D will be better, but if you never print bigger than A4 (or the slightly smaller 10x8, for those of you imperial die hards) then you don't need 15mp anyway. But then again, there are some of us who like to brag that they have the latest. Those in the know, actually recognise that the latest is not always the best.</p>

<p>My apologies for getting your name wrong Arash, I must be more dyslectic than I thought.</p>

<blockquote>

<p>If you know the stuff why are you asking any way?</p>

</blockquote>

<p>I believe you misunderstood my question. What I asked was: "how many extra pixels do you need to compensate for a given amount of extra per-pixel noise". This is not the same as the question you tried to answer, which is "how much extra per-pixel noise do you get if you increase the number of pixels on a sensor by a given amount". Part of the difference is the nonlinearity that I suppose, but not know, is introduced by human perception.<br>

My idea that read noise is important sometimes too comes from <a href="http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html">Emil Martinec's measurements</a> . They seem pretty solid to me. I never bothered to calculate how many photons from a shadow hit a pixel. I'll do that sometime when I'm bored, but not now.</p>

<p>Allard (getting pretty bored with this thread too, but won't promise to quit until I really plan to)</p>

 

<p>Arash Hazeghi exellent shot of Cormorant in flight. So this is what happens when you actually go out and take pictures of real things with a 50D.</p>

<p>This is simply a case of too much of even a good thing can be bad for you. Too many pixels that are too small packed into an APC sensor equals too much noise period. There is only so much you can fit into this space. </p>

<p>The upward limit before noise starts getting out of hand on a APC sensor is about 8-10MP. This is why I still shoot 30Ds when I want to use a crop sensor. I shopped the 40D and even it had significantly more noise than my 30Ds.</p>

<p>If I need more resolution than my 30Ds can provide I go to one of my FF bodies which has enough real estate to allow the 12 or 16 MP count.</p>

<p>High Mega Pixel count numbers are for marketing people not for photographers. We don't want the highest number of MP we want the RIGHT number of MP. </p>

<p>In the amateur market the only thing that matters to them is the largest number of MP and with lenses the only thing that matters is sharpness. So, that's what they focus on. </p>

<p> </p>

<p>More pixels means more infromation and that translates into more detailed images. The more pixels you can fit on the sensor the better the images will be. There maybe a physical limitation as to how many you can fit but I am sure that we are far from or even close to how many can be put on a sensor The retina in a human eye has so many pixels that the image is three demensional. and I am sure that someday some one will manufacture a sensor that will reach or get close to that point. Pixels are going to go up and not down that is for sure.</p>

<blockquote>

<p ><a href="../photodb/user?user_id=4709099">Michael MacDonald</a> , Aug 30, 2009; 03:20 p.m.<br>

...The retina in a human eye has so many pixels that the image is three demensional. and I am sure that someday some one will manufacture a sensor that will reach or get close to that point....</p>

</blockquote>

<p>The retina in a human eye also has a twin, the retina in the other human eye, as they usually come in pairs, it is this that allows the brain to percieve 3 dimensional depth. Although there have been some 3D film cameras they usually require special viewers and have either crude reflector prism or limited coupled prime lenses rather than zooms etc.<br>

A stereo digital camera would be an amazing breakthrough, but still some way off I fear, as sensor still have problems with peripheral light acceptance at acute angels and so forth.<br>

I tried a 50D with a 200mm f2.8L and didn't like it. No graphs or charts but I prefered -and as such still use- the images from my humble 400D. Don't really need the speed or I would have got a 40D.<br>

Might see what the 7D offers, in the main, if it offers 25fps PAL video and a minijack input for a beachtek XLR adaptor then canon may have a sale.</p>

<p>Um. As far as pixels in the retina goes... some years ago my son (who is doing well now, thank you) had brain cancer, and following several neurosurgeries, his vision was entirely out of kilter - he had one eye looking out and up, the other looking out and down, and couldn't 'put' the visual field together. Although his eyes lined up to nearly normal again, it took almost a year for his brain to knit together the visual field so that he had a fairly normal spatial vision again. The eyes are not cameras, but marvelous light-gathering organs that feed the information to the brain. The brain 'sees' - the eyes only talk to the brain.<br>

And the noise in my 50D was annoying at first... a fair amount of pattern noise even at lower ISO... but it still delivers a lot of visual information.</p>

<p>I have heard that 50D apart from its high noise gives softer images (compared with 40D). This is what I heard. Who wants to pay more to buy a camera which gives worse result? I am waiting for 60D.</p>

<p>50D is soft and noisy, no doubt about it. It's a stinker.</p>

<p>I have owned the 20D, the 40D, and now the 50D, and I can honestly say that in my humble opinion, the best of the bunch was the 20D. The 50D delivers soft noisey images that mean I have to spend time sharpening in Aperture. As a semi-pro, I take lots of images of children and families, and I can't tell you the number of times the images have been disappointing. So much so, that I have actually just bought a second hand 30D. Sure, I miss the large display that the 50D has, but I feel more relaxed on shoots because I'm more confident that my images will be satisfactory.<br>

I din't understand all the technical stuff above, but from a user's perspective, I expected a lot more from a $1500 camera than I got from the 50D. Lesson learned: When it comes to cameras, it isn't always good to be an early adopter! I think Canon peaked at the 20/30D in the amateur/semi-pro market.</p>