[Dubious Drewski]

I took some time to directly compare my old k10 with a borrowed k7. While editing photos with the k7 for the first time a while back, I felt that the images looked exceptionally noisy. I wanted to empirically determine how much the ISO performance differed.

Take a look at what I found:

Both cameras were reset to factory settings and placed in full manual on a tripod, using the same lens and identical exposure settings. I shot RAW DNG and did no editing at all, except the resampling.

It looks like at 100%, the k7 is definitely noisier than the k10, but if you resample the k7 down to the k10's resolution, it's far better.

And one thing that I find weird: the DNG RAW files from the k10 are 16.2 Mb. The DNG RAW files from the K7 are 16.3 Mb. So the K7 must be using higher compression. That would be ok if it was lossless, but doesn't the grain in these sample images look alot like an over compressed Jpeg? Are these RAW files actually using lossy compression?? I sure hope not!

If there's anyone out there who's concerned about the K7 noise performance, this should be interesting to you.

---

Last edited by Dubious Drewski; 08-29-2009 at 11:26 PM.

[Jasvox]

Nice graph. Have you spent much time comparing the low light/higher ISO photographs from the K-7 and a decent lens? I have and on average, they can be pretty remarkable. Not necessarily FF characteristics, but have seen some at 800-1600 that have been very clean, creamy and film like. I own the K10D and almost never shoot at anything over 800 with it. I can live with the noise characteristics of 800, depending on the style of the photographs being made and at times wish I had the confidence to go higher up, but until I get something like the K-7, I will stick to a max of 800. Again, the examples I have seen, in capable hands with the K-7 have been convincing. Pixel peeping aside, which isn't really photography to me, is just looking for faults.

Jason

[Dubious Drewski]

I admit that these samples were taken from a dark area of the image, meaning they're underexposed. But they're all equally underexposed, so it's ok. I'm not completely put off by this noise performance, but I do wish it were better. It's the only gripe I have with the K7 and Pentax in general.

I don't mean to insult you, Jasvox, but you say you do not shoot over 800 - well that's alright, but I need to. While shooting weddings in dark chapels where flash is not permitted or in Hockey rinks with fast action, I am forced to go up to iso1600 very frequently with my K10. Anyone who says that high-iso performance is unimportant obviously has little experience in photographing events. You often have no choice, even with glass that's f2.8 or faster.

And Pixel peeping is important if you are creating these images for a client who is demanding quality. I shoot alot, for many different types of clients - and they get on my case if everything's not perfect. I've had comments about the noise in my images in the past...

EDIT: but with this griping aside, I do love the Pentax system, and it sure has allowed me to get some great images in the past.

[Jasvox]

You are indeed correct, I don't shoot weddings or hockey matches, but I do shoot in low light locations quite often as my job in tourism demands it. 800 is my preferred limit, but that's not saying I don't go higher, which I eluded to.

On the other hand, if I did shoot action or weddings quite frequently and relied on higher ISO settings as a norm, I'd more than likely be shooting full frame anyway.

At any rate, I think the K-7 examples have been pretty promising. I am looking forward to picking one up at the end of the year when I am back in the States.

Jason

[kittykat46]

Both cameras were reset to factory settings and placed in full manual on a tripod, using the same lens and identical exposure settings. I shot RAW DNG and did no editing at all, except the resampling.

It looks like at 100%, the k7 is definitely noisier than the k10, but if you resample the k7 down to the k10's resolution, it's far better.

And one thing that I find weird: the DNG RAW files from the k10 are 16.2 Mb. The DNG RAW files from the K7 are 16.3 Mb. So the K7 must be using higher compression. That would be ok if it was lossless, but doesn't the grain in these sample images look alot like an over compressed Jpeg? Are these RAW files actually using lossy compression?? I sure hope not!

If there's anyone out there who's concerned about the K7 noise performance, this should be interesting to you.

RAW means the original data captured by the sensor hardware is intact, it doesn't necessarily mean NO data compression. Given the potential unmanageable size of large MP RAW files, I won't be surprised if work has been done on loss-less compression.

Resampling the K-7's RAW file to 10.2 MP definitely involves data interpolating - that's basically like JPEG downsizing - so its not surprising if the grain looks smoother.

You have actually stumbled on one of the often overlooked advantages of shooting with a Higher MP sensor. IF the inherent noise from a 14 or 15MP sensor does not get any worse (that's a big IF), downsizing the captured image to 10MP can result in better Image Quality - more detail, less grainy than one captured directly from a 10MP sensor.

I noticed this years ago with shots from a 6 MP sensor (which was high-end state-of-the-art then) vs. shots from a 3 MP sensor.

[pcarfan]

Which RAW converter did you use?....I think RawTherapee (Free) which can be set so that there is no default NR and sharpening done to the images is the only unbiased way of comparing two cameras (AFAIK, you can't do so with any of the adobe products). But, I think your test is still valid. I've done all my 'testing' with lightroom...so, I am not finding fault....I do the same thing....I was just mentioning something I learned recently, and I downloaded RT two days ago, and will be using that for high iso images from now on.

These are just my thoughts, please correct me if I am wrong. As I didn't do this actual test. But, my feelings were that at iso 800 or there abouts, this is exactly what you will see (I pretty much said the same thing as your observation shows, in a discussion in DPR a couple of days ago) The fact that K-7 shows better performance when re-sized, and your desire for print quality means the k-7 is still much better for printing purposes at iso800.

I think at iso 1600 you will find the K-7 to perform much better (If you can post your results). I am afraid that at iso 100 and 200, the K10D may fare even better, especially if the iso 100 shot is underexposed and then recovered in PP (It is only my hunch, I could be wrong, even if I am right,I don't think there will be any implications in real-life, as any advantage will be minute).

I have both the K10D and the K-7 and I may run these tests, if I get some time.

Having read what I had written, mostly may assume that the K-7 is not much of an improvement over the K10D for iso performance. One could not be far from the truth. It is just day and night when it comes to real life usage.

AFAI am concerned, Pixel peeping is very important for testing purposes to understand and critically evaluate performance. It has helped me tremendously with actual picture taking, when I understand everything at a macro level. However at times things cannot be directly translated form the macro to the field.

At high iso (1600 and above) the K-7 is remarkably better. Also, the K-7, even though noisy retains the IQ to a very high degree all the way to iso 3200. In reality, where PP is involved, the K-7 images at high iso with PP is highly usable as it retains IQ, where as the k10D to me are not. So, in real life I will gladly use the K-7 to iso 1600, where as I only used the K10D to 400. I can stretch the k-7 to iso 3200 and the k10D to 800. So, to me there is a clear two stop advantage. This is comparing them at pixel level. If the K-7 images are downsized the improvements are even more (this matters for printing....).

[Digitalis]

I was never particularly impressed with the K10 at anything above ISO 400. hence the reason why I have invested in the fastest lenses I could get. Pentax often say their cameras rendering is "film like" really, that's a pretty easy promise to keep.

---

Last edited by Digitalis; 08-31-2010 at 05:22 PM.

[mithrandir]

And one thing that I find weird: the DNG RAW files from the k10 are 16.2 Mb. The DNG RAW files from the K7 are 16.3 Mb. So the K7 must be using higher compression. That would be ok if it was lossless, but doesn't the grain in these sample images look alot like an over compressed Jpeg? Are these RAW files actually using lossy compression?? I sure hope not!

The PEFs for the K10D are compressed, the DNGs are not. DNGs on the K7 are now compressed with lossless compression. This is one reason why I shoot in PEF on my K10D and convert to DNG in Lightroom.

[Urkeldaedalus]

Interesting findings. This means that if you know you want to print big it's still wise to go with 14.6 mp, but if you know it's going to be 8 X 10 or smaller (or posted on the web) you can use 10.2 mp. Nice to have that flexibility.

I have a question -- I'm assuming when you all are talking about "downsizing" the images you mean doing so after the image is shot either in camera or with software like a RAW converter. Given this, is there any benefit of just shooting at the smaller megapixel size with the original picture, or does it have to be downsized afterward to see this positive effect?

[Ash]

I'm not sure it will make too much difference if you downsize a 14.6Mp image to 10.2Mp or just shoot in JPEG 10Mp mode in camera. I'm also uncertain that a downsized 14.6Mp image will give any less appreciable noise on a 8x10 print than if a full 14.6Mp image is used for the print. The full sized image will be resized anyway to shrink enough to fit to the print format. Noise will subsequently be downsized accordingly.

Either way, shooting at full-size, you'll see noise much more than when downsized in any way.

[Marc Sabatella]

It looks like at 100%, the k7 is definitely noisier than the k10

Well, sure, because you're blowing it up bigger. Consider - take one image from the K10D and view it at 100% in one window but at whatever size you need to view it at to emulate the K-7's 100% in another window (maybe 125%). You're blowing it up bigger, so everything - including the noise - looks worse. But obviously, it's not an inherently noisier image - it's the *same* image.

Comparing images at different sizes is always a highly dubious way of comparing noise, at least as it affects images in the real world.

[Dubious Drewski]

Interesting findings. This means that if you know you want to print big it's still wise to go with 14.6 mp, but if you know it's going to be 8 X 10 or smaller (or posted on the web) you can use 10.2 mp.

Just so you're aware, 10Mp will allow for much bigger printing than a simple 8x10, as long as the pixels are clean and accurate enough. Take a look at this image, for instance. It's an extremely cropped portion of the original image, leaving it at 2.6 Mp, yet it's sharp enough that I'm able to print an 11x17 print that is absolutely gorgeous even when viewed up close. (It works out to be 165 dpi)

It's a matter of quality versus quantity of pixels - you can print larger more successfully with 2.6 high-quality megapixels than you can with 10 megapixels of mediocre quality. And that's the truth.

It looks like at 100%, the k7 is definitely noisier than the k10

Well, sure, because you're blowing it up bigger. Consider - take one image form the K10D and view it at 100% in one window but at whatever size you need to view it at to emulate the K-7's 100% in another window (maybe 125%). You're blowing it up bigger, so everything - including the noise - looks worse.

You're ignoring the fact that I also included a downsampled version of the K7's image. I did this to address the very issue you are bringing up here. When you "zoom out" from the K7 image (ie: downsample it to 10.2 megapixels), it turns out to be much cleaner than the k10.

You have actually stumbled on one of the often overlooked advantages of shooting with a Higher MP sensor. IF the inherent noise from a 14 or 15MP sensor does not get any worse (that's a big IF), downsizing the captured image to 10MP can result in better Image Quality - more detail, less grainy than one captured directly from a 10MP sensor..

There was no stumbling involved - I knew that the IQ would increase with the downsampling. That's why I did it in the original image.

[thirdofthree]

You're ignoring the fact that I also included a downsampled version of the K7's image. I did this to address the very issue you are bringing up here. When you "zoom out" from the K7 image (ie: downsample it to 10.2 megapixels), it turns out to be much cleaner than the k10.

But downsampling includes processing, so you're comparing two very different animals:

<1> K10 @ 10.2 MP
<2> K7 @ 14.6 MP > downsample > K7 @ 10.2 MP

Given this comparison, it's unclear whether the cleaner K7 @ 10.2 MP image is a result of better ISO performance or simply the downsampling process.

[pcarfan]

But downsampling includes processing, so you're comparing two very different animals:

<1> K10 @ 10.2 MP
<2> K7 @ 14.6 MP > downsample > K7 @ 10.2 MP

Given this comparison, it's unclear whether the cleaner K7 @ 10.2 MP image is a result of better ISO performance or simply the downsampling process.

So, what is your suggested methodology?

[thirdofthree]

So, what is your suggested methodology?

Heh, I figured someone would ask. To be honest, I'm not competent enough in the realm of digital photography technology (yet?) to detail a proper methodology. I was hoping some of the far sharper and more experienced PFers could chime in. Sorry.