10D noise pattern

[sean_harding]

<p>I've noticed a somewhat strange pattern in the noise in high ISO images from my

EOS 10D. I haven't yet seen it as noticable in prints, so I'm not terribly concerned

about it, but I'm curious if anyone else has seen thise (or if, perhaps, I'm just

imagining things). Here is a sample image:</p>

<img src="http://www.sharding.org/outgoing/temp/noisecomp.jpg" width="450"

height="200" alt="Noise comparison" border="2" />

 

<p>That image contains three samples. One is a JPEG, one is RAW converted in

Photoshop CS and one is RAW converted by Canon's File Viewer Utility. The RAW

samples are from the same original file, the JPEG is a different file. Exposure was

identical for all images, default conversion settings were used and no manipulation

was done in Photoshop other than that needed to combine the three images. The ISO

setting on the camera was 3200, to maximize the effect, but it's visible at lower ISOs

too.<p>

 

<p>Basically the thing that I'm wondering about are the little horizontal and vertical

lighter lines in the image, especially in the RAW to Photoshop version. I have arrows

pointing out two instances in the first image. The top arrow is pointing at a horizontal

line and the bottom arrow is pointing at a "backwards L" shape. They look kind of like

little pieces of rice.</p>

 

<p>This is definitely more pronounced in the Photoshop photo, so it seems that

something about Photoshop's RAW conversion is contributing to it. However, the

pattern is still somewhat visible in the FVU photo (the horizontal mark is in both

samples), so it's not all Photoshop's fault. Like I said, this doesn't seem noticable in

any of the prints I've made -- I've only seen it when viewing on screen at 100%. So it's

not a major problem for me. More than anything, I'm curious about what is causing

this. It's not like the noise I've seen from any other digital camera I've had.</p>

[mark u]

What aperture were you using? It might be sensor dust.

[sean_harding]

The sample was shot at f2.8 (70-200 2.8L IS). However, from what I've seen, there's

no correlation with aperture -- it's visible in every high ISO shot. It's also basically the

same throughout the image in every image and has been since I've had the camera, so

I don't think it's dust. But I'll clean the sensor and take some more test shots to rule it

out.

[william_le]

Very interesting to know. Thanks

[andreas_carl]

I doubt it is dust - looks like random noise to me. Patterns occur by accident in any random distribution. For example if you throw a dice aa thousand times, you will get six sixes in a row from time to time...

[peter_foiles2]

It is deinitely NOT dust. You will see these kinds of artifiacts with just about any digital camera if you look hard enough. They are much more prevalent in files from my Oly 5050 and Minolta A1 than my 10D but as you see they can be found there as well. Someone with more technical expertise will have to explain exactly what the artifact is due to.

[paul_nicol]

15 years ago I wrote some graphic programs that would convert pixels on screen (really an address in memory) to print patterns of gradient color on a printer. Not being a mathematician I tried to come up with some crude algorithms that would try and keep the patterns random in appearance. I wasn�t always successful as when I would get a particular �byte-word� pattern it would set off my program to generate similar types of lines (or patterns) you are seeing. It would fix itself once it found a different color pattern on the �screen�. I�m guessing even the real mathematicians have a problem with this stuff and that is, more or less, what you are seeing. Kind of falls into the same thing Andreas brought to light. I don�t work at graphics anymore so don�t even bother with the stuff as what you can buy now is streets ahead of what I was doing. As you said you can�t see it anyway. My two cents.

[alex_culiuc]

When you run NeatImage with strong noise-removing setting on a VERY noisy image, you end up with ABSOLUTELY the same results. In NeatImage it can be even more pronounced -- imagine an absolutely smooth skin (smoothed by NeatImage) with such streaks and lines appearing on cheeks and on the forehead -- on screen, at 100% they look quite disturbing.

 

My guess is that you ran the PhotoShop RAW converter with some sort of noise-removal featured activated (this is quite obvious -- the PhotoShop RAW version looks much cleaner than the Canon RAW version).

[qiang_lin]

I think Alex is right. With finite number of samples (pixles), random noise is not really random with zero mean as we assumed. Any filtering will make this (non-zero-mean) more noticable in a form a another because the result is less noisy.

[arunas_salkauskas]

<p>There's a colour filter over each pixel on the sensor.</p>

 

<p>so you have a pattern of <br>

grgrgrgrgr<br>

bgbgbgbgbgb<br>

grgrgrgrgr<br></p>

 

<p>At high ISO, the signal to noise ratio is low, so suppose one of your Green pixels (a G) gets a few extra counts of noise. the Raw conversion will interpolate from that one to the nearest green neighbour on either side, resulting in a slightly brighter line for at least 3 pixels. If there's any sharpening or edge enhancement going on then this brigher line will get enhanced too. This is a fairly simplistic view of the interpolation - hopefully they do a better job than this, but you never know. </p>

 

<p>But frankly, you're shooting at ISO 3200 - there's gonna be noise. </p>