Digital Cameras and their noise

November 2, 2003 2:34 PM

This excellent overview of pixel noise in digital images also offers up some plugins to help make your photos look better and print cleaner. I'd noticed the effects and noise in much of my digital photos, but had no idea how to remove it -- until now.
posted by mathowie (22 comments total)

Perfect for those those low-resolution free-porn sites.
posted by mischief at 2:37 PM on November 2, 2003

Apparently bob is angry with us (scroll down to the first comment in the link).
posted by The Michael The at 2:55 PM on November 2, 2003

On Soviet Metafilter, Red Hot Anal Slut is You!
posted by Stan Chin at 2:59 PM on November 2, 2003

This is definitely useful for me. I have the same camera as him, and have noticed the same problems in the images. They look fine on the web, but the few that have been printed out look kind of shitty.
posted by angry modem at 3:03 PM on November 2, 2003

*Waves at Bob in the cheap seats*

[Useful link Matt...thanks]
posted by i_cola at 3:05 PM on November 2, 2003

I have noticed the Mae problem with digital cameras and, being an extreme PhotoShop novice, usually make things worse when I try to fix them. This article is a little advanced for me, but gives me hope that there is a solution and I have filed it away for future reference.
posted by dg at 3:07 PM on November 2, 2003

I guess my eye isn't discriminating enough, as I can't really such much of a difference between all the image examples provided. Then again, I'm not as much of a purist as, for example, my roommate yo, kevspace! who rarely watches movies compressed with DivX because it's sacrilege to the original DVD imagery. Or something like that.
posted by WolfDaddy at 3:20 PM on November 2, 2003

Wow. This dude really prefers those smeary-blurry airbrushed-looking versions over the detail ones? Different eyes for different folks, that's for sure. Then again, in the modern world everyone's a photo expert after spending a few weeks with a digicam and photoshop.
posted by HTuttle at 3:25 PM on November 2, 2003

The uber-God of comparisons of image noise reduction tools is Michael Almond's roundup. Although there are one or two he doesn't cover, most of 'em are in there.
posted by kindall at 3:28 PM on November 2, 2003

Interesting stuff.

Oh, and Hi, Bob! Too bad about the registration stuff and all.
posted by Ayn Marx at 3:36 PM on November 2, 2003

ehe. now you can blow up those cat schwartz pictures extra big.
posted by carfilhiot at 3:49 PM on November 2, 2003

Too bad about the registration stuff and all.

no pooty for bobby's booty!
besides, stan chin will never share.
posted by quonsar at 3:53 PM on November 2, 2003

I think this tutorial and the associated actions (in the conclusion) have done more for my post-process workflow than any other single thing.
posted by cedar at 4:14 PM on November 2, 2003

What HTuttle said. I prefer the originals. I'd like to see a widely-usable image format that lets you store that kinda filtering info as metadata, so you don't loose any of the original info.
Ditto for audio. I'm sort of working on that myself, putting normalizing info in the availible metadata areas of FLAC files, then normalizing them on-the-fly as they're accessed via samba. Not quite there yet.
posted by duckstab at 4:47 PM on November 2, 2003

I've discovered, over the years, that photoshop's print engine does not always yield the best results. For windows people, take a hard look at Vueprint Pro and compare it's results with Photoshop or, whatever you use. You'll be suprised.
posted by Dean_Paxton at 6:44 PM on November 2, 2003

angry modem i also own a canon G1 and was very surprised when it produced worse images then my old s10 (gotta love those canons). So far i haven't been doing much printing but i can say that when i finally stopped cheaping out and purchased some high quality paper the pictures came out much better. I still much prefer the images on the computer then on paper. Thanks matt for this very useful link.
posted by NGnerd at 7:13 PM on November 2, 2003

I just tested this, and it worked pretty well correcting the digigrain my Canon G2 dropped in a night-time long exposure shot. I used the Level 3 correction -- Level 9 is probably more than necessary for most shots.
posted by me3dia at 8:40 PM on November 2, 2003

I don't really like the result of this guy's actions -- the photo comes out with that crackly-synthetic-speckly look that I associate with oversharpening. But to each his own.
posted by argybarg at 9:53 PM on November 2, 2003

Yeah, I must say I find the way he looks at images somewhat strange. He laments the amount of noise in his first picture and says the issue is minor with the other two. However, I think his "fixed" version of the first picture is noticeable worse, while the 2nd and 3rd examples seem similarly improved by whatever it is he did.

But then, this is all a curiosity for me, I just look at good photos - I don't take them.

Lastly, Htuttie, read the flipping page. 25 years photography experience and 10 years photoshop experience beats "a few weeks" by a couple orders of magnitude.
posted by kavasa at 5:35 AM on November 3, 2003

rarely watches movies compressed with DivX because it's sacrilege to the original DVD imagery

You should mention to your friend that DVD's are actually compressed data as well. I've found that watching DVD's on really large plasma displays can be a horrendous experience if you're sitting too close, due to fast-motion artifacting. As for DivX's, well, if you want it to fit on a CD, you gotta make sacrifices.

My only recommendation to our guide writer would be to shell out more money on the right kind of gear if he wants good results. For instance:

For a long time I blamed the issue on my 600 dpi printer, until I upgraded to a 2400 dpi printer and saw no improvement whatsoever.

This is bad gear to buy if he wants to improve his images. Pixels are pixels are pixels. If you've only got 2000x1500 pixels to work with at 180dpi (the resolution out of the camera), it won't make a difference whether your printer is printing 5x7's at 600 or 2400 dpi since the printer is already interpolating to adjust for the lack of data. A 2000x1500 (at 180dpi) image translates to 5x7 prints at 300dpi, or 3.5x2.5 at 600 dpi. You simply cannot print data that isn't there.

His noise reduction algorithm is decent enough, but I wouldn't let it near any images I wanted to print. The reason is in the details, or rather, the lack thereof. If you open the beach scene images (before and after) and look at the rocks and plants to the right, you'll notice that his filter has killed whatever crispness the highlights offered. For web images, this isn't a problem, but a 2400dpi Epson print on nice paper is going to look more muted than the original. Unless you've got noticable hotspots (like from long exposures) that are easily Photoshop'ed away, or color corrections that only change the hue of the pixels, not the data itself -- then leave the image alone!

If he's unsatisfied with the quality -- well, you get what you pay for. Shell out a grand for a dye-sub printer that will look better interpolated, or better yet, buy a better camera.
posted by Civil_Disobedient at 10:24 PM on November 3, 2003

I checked the original of the first picture several times. That is not noise, it's real detail. Noise is random, but the skin detail clearly looks like real skin. Anybody who thinks otherwise needs to stand in front of a mirror with a really good light and really look at their skin. It's hardly the infinitely smooth stuff of 3D video games.

If you don't like the detail that's fine; there's taste to photography. But you really shouldn't call it "noise"; it's offensive to the camera designers ;-)
posted by Jeremy Bowers at 9:40 AM on November 4, 2003

it won't make a difference whether your printer is printing 5x7's at 600 or 2400 dpi since the printer is already interpolating to adjust for the lack of data

Well, in reality, printer dots don't map directly to pixels. Each printer dot is either on or off, while a pixel can vary in intensity; you thus need numerous dots to reproduce a single pixel. This is typically done using halftoning on a laser printer or diffusion dithering on an inkjet. So a 2880 DPI printer may reproduce better detail than a 600 DPI printer even when the source is only 180 DPI. The exception to this rule is a dye-sublimation printer. This type of printer can actually vary the intensity of its dots and so 300 DPI is really 300 DPI. Regardless of the type of printer, though, you're not going to get 16.7 million colors out of it; if you look at the number of levels supported by the halftoning or dithering or dye-sub hardware, it's just not mathematically possible.

Confusing the issue is the fact that the DPI figures used in marketing inkjet printers are basically bullshit anyway. If a printer has "2880 DPI" resolution, that means it can position dots with a precision of 1/2880th of an inch. It does not necessarily mean that the size of the dots is 1/2880th of an inch. In reality, if you printed every other dot on such a printer, you'd get a solid line (rather than a microscopically dashed line as you might expect) because the dots overlap -- they're significantly bigger than implied by the DPI figure. Being able to position dots as precisely as possible is good, but making the dots smaller is better. Being able to make the dots vary in size would be better still, and some inkjets are beginning to do that.

BTW, as a rule of thumb you want 150-300 DPI in your source image regardless of the DPI of your output device. (The more the better up to a certain point, which is about 300 DPI. Beyond that our eyes can't distinguish additional detail anyway.) The 5 megapixel prints I make to US Letter papers using my Canon inkjet printer work out to exactly 256 DPI, which is a nice round number in hexadecimal and thus pleasing to my inner geek.
posted by kindall at 3:00 PM on November 4, 2003