Nikon D3000?

[extra]

Why are you resizing the crops to make the D40 look all blurry?

Because comparing images at the same actual size is the only thing that makes any sense. (disclaimer: throw the same magnification of sensor size in there as well, but in the case of this thread for aps-c dslrs it's all approx the same give or take a tiny bit).

Let's say you have a 7d shot in one hand a D40 shot in the other hand. Now, say you look at the at the image at 100%. Imagine that *per pixel* the D40 is sharper (it won't be, but let's assume). If you are going off of the "pixel level" image quality reference that you said matters to you then you might conclude that the D40 has better image quality. This would not only be the wrong conclusion but would in fact also be completely batcrap insane!

This would be the same thing as doing this:
1. I walk into a store and buy two rolls of 35mm film. I buy a role of Velvia 100F and a roll of that Kodak Ultra Max 800 film.
2. I take some pictures and have them printed. I want to compare the image quality between the two films. The 800 speed film has big grains (think: less megapixels) and the Velvia 100F has small grains (think: more megapixels).
3. I see this and think okay, well we'll print them based on the grain size (think: viewing the image at 100%). I print the Kodak film at 4x6 and turn the Velvia into a 18x24.
4. I then look at the pictures and conclude that the films have about the same image quality.

Insane, right? Yet this is what we are doing when we compare cameras with the same size of sensor but different megapixel counts at 100%. Instead, we need to compare images at the same actual size (think: taking two 35mm negatives and printing both at the same size).

Upping the size of the smaller image is the best way to do this. It does not introduce more noise. If done right it does not hurt the image quality. It does not remove details from the image. You could downsize the larger image but doing that tends to remove detail--and some argue it sometimes tends to have a sharpening effect on images--still, it is sometimes useful to do that for comparisons.

To take the argument to a ridiculous level: Say Nikon introduces a D9000 extreme with 90megapixels. Say at the pixel level the image is 3x as noisey as that of a D40. I swear some people here would say that the D40 had better image quality. It would happen.

And yes, if you are doing only screen viewing then viewing images at 100% is even less relevant than in prints. The size of your screen does not magically change when you get a higher megapixel camera folks.

Here is a write-up by a very smart guy explaining why more megapixels is a good thing: http://photography-on-the.net/forum/showthread.php?t=747749 . I highly recommend reading it.

Ken Rockwell's site has some good info (for example he tells you to just go out and take pictures, get into it--best advice) however some of the stuff he says, like about the D40 being teh best camera evah, or his love affair article for one of the Canon digital elfs also rings a bell lol, are often really nuts and/or flat out innacurate as hell. This does not mean his site isn't great. He actually comes right out and tells you that his stuff is for entertainment purposes and should not be taken seriously--he comes right out and says he does a lot of it for fun. Seems like a great guy.

Now, one last thing. Everyone knows the 7d's iso 12800 is horrible right? The reviews say it's for emergency use only, right? Looks horrible at 100% that's for sure! Well here's an iso 12800 test shot from last night, processed with my one-click photoshop 7d noise reduction action. It's re-sized to be equivalent to a 100% crop from a D40 (it was re-sized to 6mp then i took a 100% crop from that):

 

[Munky]

I disagree. You're trying to group two completely different media, and that makes absolutely zero sense to me. Print and screen viewing are not comparable, because a high quality print is of much, much higher resolution than any computer screen. People looking at pictures on a monitor only view them at 100% or smaller sizes. People making large prints are the ones doing the enlargement. You're trying to simulate a printed enlargement by resizing crops on a low rez monitor, which only serves to blur the image.

So, if you wanted to fairly compare cameras at the same enlargement, you'd have to actually print the photos and look at the results, not simulate printing by resizing. On the other hand, if my target media is an online gallery, I would be resizing the photos to make them smaller, not larger, to fit the resolution of typical monitors, and in that case I'd care about per-pixel detail a lot more than the number of megapixels in the image.

 

[ElFenix]

i wish dpreview would print at the same size and compare that way rather than do the equivalent of printing the 7D at 12 and a half square feet and declare it worse than the D300 examined at 8 and a half square feet. which is what per-pixel comparisons are.

unfortunately printing doesn't show results for everyone to judge unless the prints are mailed all over the place. and magazines are nearly dead, so it's no way to run a new business. viewing images onscreen are the only way to realistically do this (of course those who are truly interested can download the samples, print, and compare. unless the samples are sloppy like on imaging-resources).

examining same-sized on screen is no different than looking at the same sized prints. larger prints aren't blurring, larger prints are revealing that the detail simply isn't there. an 8x10 contact print from a large format will have more detail than an 8x10 35mm. enlarging the 35 mm film isn't creating the blurring.

per pixel detail of the resulting image of a downsized image can be sharper with a lower-sharpness per pixel but higher res sensor than with a higher sharpness per pixel and lower res sensor. see the first paragraph again for why per-pixel comparisons are invalid.

 

[Suspicious-Teach8788]

now that's interesting. so nikon took a step back from the D60 with the D3000?


edit: did the SD880 really hit $1000 as rockwell is claiming?

Something like that. The SD880 when I wanted to buy was like $300. I bought a SD990 instead. Then a few months later it was $500. The SD880 is just THAT good. I think it's a simple supply & demand though. They're so rare because they're not being made that they're being marked back up to like MSRP or something. Or higher...

But it's sad the D3000 was a step backwards. On paper it was awesome. It was an awesome upgrade to the D40 in all respects... and then the image tests came out and I barfed. But luckily the D5000 is a hit.

 

[twistedlogic]

examining same-sized on screen is no different than looking at the same sized prints. larger prints aren't blurring, larger prints are revealing that the detail simply isn't there. an 8x10 contact print from a large format will have more detail than an 8x10 35mm. enlarging the 35 mm film isn't creating the blurring.

You post this and then link to an article that says:

"If you do equal-sized prints from the two cameras you will not find that DPReview's noise comparison is right, you will find that this one is. Yes - DPReview, and your monitor have indeed been lying to you for years. "

Ironic isn't it.

What is wrong with out of the box testing? Do you expect the average user to immediately take it and twist the settings to produce something else?

Posted by Daniel Browning: (cause he sums it up better than I do)

"Out-of-camera images.

Processed formats that come directly out of the camera, such as JPEG are good for drawing conclusions about the utility of that processed format for whatever purpose is needed, but it probably does not accurately reflect the sensor itself. Furthermore, any conclusions, which are necessarily subjective, cannot be generalized to pixel sizes in all cameras. Too much processing has already been applied to the raw data, including noise reduction, saturation, black point, tone curve, and much more, which all have an affect on apparent noise, sensitivity, color, and dynamic range."

Anyway, I'd think that a user staring at Dpreview %100 image crops @ high ISO's and posting on camera forums aren't your typical "average" user.

Let's say you have a 7d shot in one hand a D40 shot in the other hand. Now, say you look at the at the image at 100%. Imagine that *per pixel* the D40 is sharper (it won't be, but let's assume).

Funny you should post that. I've been reading many complaints that the 7d is soft due to the AA filter being so strong.

 

[ElFenix]

You post this and then link to an article that says:

"If you do equal-sized prints from the two cameras you will not find that DPReview's noise comparison is right, you will find that this one is. Yes - DPReview, and your monitor have indeed been lying to you for years. "

Ironic isn't it.

not when you realize it's comparing two images from different cameras with different resolutions on the monitor at 100% pixel for pixel that he's complaining about. which is what i've been complaining about.

context.

 

[extra]

You post this and then link to an article that says:

"If you do equal-sized prints from the two cameras you will not find that DPReview's noise comparison is right, you will find that this one is. Yes - DPReview, and your monitor have indeed been lying to you for years. "

Ironic isn't it.

Funny you should post that. I've been reading many complaints that the 7d is soft due to the AA filter being so strong.

I don't see how that's ironic at all lol. He's saying that looking at the images at 100% on your screen has been misleading you this whole time. Which is exactly what we are saying.

And the AA filter thing is funny. You know when the 40d was released people coming from the 20d/30d said exactly the same thing, right? And indeed, it's actually true! The 7d raw files with no sharpening applied at look a little soft when viewed at 100%! When processing the raw files in DPP, images from the 7d take one or two numbers higher of sharpening than images from a rebel XSI.

There is a good reason for this and the "softness" in this case has absolutely nothing to do with image quality or amount of detail captured! The 7d captures more detail than the XSI! I suggest reading up on what the AA filter actually does and why sharpening is required.

Comparing unsharpened raw files is naive at best. Here's a great example for you: Grab a nice raw file from a Leica M9 (which has no AA filter) and a Nikon D3x. Process them both with absolutely no sharpening. Wow, the D3x looks horribly soft, the Leica is so much sharper, amazing! The D3x's image quality must be worse than the Leica!!! Yet, what happens when you use capture sharpening on the D3x image. It sharpens right up and shows that it captures just as much detail as the M9. What happens when you apply capture sharpening to the Leica? It doesn't bring out any more detail.

 

[magomago]

You post this and then link to an article that says:

"If you do equal-sized prints from the two cameras you will not find that DPReview's noise comparison is right, you will find that this one is. Yes - DPReview, and your monitor have indeed been lying to you for years. "

Ironic isn't it.



Posted by Daniel Browning: (cause he sums it up better than I do)

"Out-of-camera images.

Processed formats that come directly out of the camera, such as JPEG are good for drawing conclusions about the utility of that processed format for whatever purpose is needed, but it probably does not accurately reflect the sensor itself. Furthermore, any conclusions, which are necessarily subjective, cannot be generalized to pixel sizes in all cameras. Too much processing has already been applied to the raw data, including noise reduction, saturation, black point, tone curve, and much more, which all have an affect on apparent noise, sensitivity, color, and dynamic range."

Anyway, I'd think that a user staring at Dpreview %100 image crops @ high ISO's and posting on camera forums aren't your typical "average" user.



Funny you should post that. I've been reading many complaints that the 7d is soft due to the AA filter being so strong.

Just to respond to my portion:
We need to differentiate a SENSOR test vs a CAMERA test. We also need to decide to WHOM the test is aimed towards.

I never said that I would use straight jpgs from a camera as representative of what a sensor can do.
For the average user, a test looking at the whole package is the key. It only makes sense because that is how it will be used: a 'functional test'.

DSLRs can be tweaked in so many ways to give you different IQ....the main reason to use default settings is that these settings are what is shipped by the manufacturer. You could sit there and subjectively tweak settings all day long without coming to a single "best settings" answer. So go with what the Manufacturer sets to default. It still isn't the perfect answer, but in an age where everyone seeks direct 1:1 comparisons, its the best option.

(To elaborate on the consumer's desire to make 1:1 comparisons....look at the people arguing about what the D300/D90 is going up against...or people telling me that the K7 is aimed to target the D5000...they can't accept that each camera company has their own setup that isn't directly correlated to other cameras...and we are just talking about simple marketing at this point, nevermind actually drawing conclusions about image quality!)

You'd be suprised how many people I know will go online to search for camera reviews and land on on DPREVIEW. They then make a camera decision based on tiny 100% crops of 'high ISO shots'. I know too many people that go on and completely read the information wrong or put the wrong emphasis on the data: "Oh noes at 6400+ ISO, it looks like the delicate feathering in this shot isn't visible anymore! I think I won't select this camera! The other camera looks slightly better at this part!".

I can't say what the best way to test camera quality is. At this point, I would argue that any test trying to gauge overall camera performance should be based upon the final use of the images. The average user, based on what I've seen and what we could probably come up with in consensus, generally prints no greater than 12x18, resizes images for their desktop wallpaper, or looks at their photos 100% sometimes when they are looking at details with friends, or even toss them on their ipod to show others.
I'd love to see reviews where, in addition to looking at 100% crops of scenes that people would REALISTICALLY take of, images get resized to some wallpaper size (lets go with 1080p since that is the new 'hd' standard) and then compared without zooming in like crazy - now we can talk about what has more naturally looking grain, and what looks like ugly noise.
It may not be the best test is searching for the minor differences between camera processing, and sure as well doesn't look at the influence of each component of the camera is but it sure as hell is a lot more honest about what the final intended use is going to be.

On that basis, the differences between DSLRs and P&S would be more minor. We'd probably see a greater emphasis on camera features more than anything else which exists in the current "HIGH ISO IQ, I see some noise at ISO800 at 200%" fest.
And it is part of why the camera market is segmented the way it is: features, ease of usability, cost, images that look fine in the above described scenarios.

 

[spikespiegal]

I suggest reading up on what the AA filter actually does and why sharpening is required.

First, spin it how you want, but USM *does not* increase detail in an image that has been lost. USM is a perceptual thing only.

Next, AA *does* cause the loss of image detail. However it's a tradeoff with the issues that non AA aquisition causes. Personally I'd like to see AA done in software with the option of turning it often. Fuzzing up an image with AA, and then sharpening it again in PP simply isn't logical.