[Class A]

I don't know if I am up to the same skill as what the Sony user posted, but this came out of a K7....

Good portrait, the face is very well lit but I personally would prefer the hair to not be that indiscriminate against the background. Image content aside, the image doesn't strike a chord with me because

• there is no place where the detail bites. Even if it is just a small area like the eye lashes or the jewelery, I like a part of to be really crisp. This is just too soft overall for me. I realise that your client appreciates that we cannot count her pores but something needs to be crisp for me.
• The softness doesn't look like thin DOF OOF. It looks like PP softness.
• I don't see as many shades as in the other shot. The transition from brighter to darker on her forehead looks almost posterized.
• The lack of colour doesn't give me the sensual feel for the skin the other shot gives me. The colour rendition might have been great but I cannot judge it with this one.
• Her mouth, chin and neck areas don't look human to me.

All this stuff is highly personal. For me your image doesn't want me to achieve the same result but that's just me. I bet your client was really happy with this, I really do and I mean that in every positive way there is.

[Jazzyboy]

Her mouth, chin and neck areas don't look human to me.

...

[falconeye]

I might be fooling myself but I've yet to see an image from the K-7 that bowls me over because of its ability to effortlessly create a sense of natural beauty like some I've seen from the Sony A900.

Forum member Torphoto (PENTAX Photo Gallery: Artist Bio - Brendan Bhagan) used to have gorgeous beauty images shot with Pentax, e.g.


This photo is also available here (requiring registration): triniPULSE.com :: Fashion :: Portfolios :: Brendan Bhagan

Unfortunately, he sold his K-7 because of its (now solved) green line problem.

The above photo is one of my all time favourites.

[m8o]

REALLY dumb ass question, are the ISO ratings taking into account the pixel account and using a base image size?

I have owned a K100D with it's super duper low light sensor and a 20D which at 100% crops looks way, way worse at ISO800 or 1600, re-sample to 6mp or upsample the 100D to 14mp and it's a draw.

I tried this line of reasoning 'bout an observation a years or few ago. Members didn't warm to the observation. ...I'm still with you however.

[rawr]

REALLY dumb ass question, are the ISO ratings taking into account the pixel account and using a base image size?

No, I don't believe so. In a manner of speaking, they aren't related to image size at all. They are tests of camera sensor performance as captured in RAW data.

From the DXO Mark site you can obtain pages and pages of technical information about how they conduct their tests and what it all means. But the low-light ISO metric ...:

Photojournalism & action photography: Low-Light ISO |

...photojournalists and action photographers often struggle with low available light and high motion, so that achieving usable image quality is often difficult when pushing ISO.

When shooting a moving scene such as a sports event, action photographers’ primary objective is to freeze the motion, giving priority to short exposure time. To compensate for the lack of exposure, they have to increase the ISO setting, which means the SNR will decrease. How far can they go while keeping decent quality? Our metric, Low-Light ISO, will tell them.

The SNR indicates how much noise is present in an image compared to the actual information (signal). The higher the SNR value, the better the image looks, because details aren't drowned by noise. SNR strength is given in dB, which is a logarithmic scale: an increase of 6 dB corresponds to doubling the SNR, which equates to half the noise for the same signal.

An SNR value of 30dB reflects an excellent image quality. Low-Light ISO is then the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9EVs and a color depth of 18bits. As cameras improve, the Low-Light ISO will continuously increase, making this scale open.

Photojournalism & action photography: Low-Light ISO

BTW the ugly table and chart at the beginning of this thread was created by me just as a way of summarising some of the DXO camera results. It was not produced by DXO.

They have a far more sophisticated and useful summary of the low-light ISO camera scores available here:

DxOMark Sensor

[Class A]

REALLY dumb ass question, are the ISO ratings taking into account the pixel account and using a base image size?

Not sure which ISO ratings you are referring to exactly, but DxOMark has two versions of each measurement graph. Select the "screen" tab and you get the meaningless pixel performance. Select the "print" tab and you get the normalised result you are talking about.

Forum member Torphoto (PENTAX Photo Gallery: Artist Bio - Brendan Bhagan) used to have gorgeous beauty images shot with Pentax,

I shall look at these, thanks.

In the image you included I don't see the "magic" I see in the A900 image. It is a great shot but it somewhat has written "digital" all over it.

There is oversharpening. This could be an artefact of aggressively optimising for the size posted, but I only have this version to look at. Also, less sharpening could address this, but will the detail still pop as it does now?

There is also less smooth gradation than in the A900 image. Could be a result of the lighting, but there seem to be a few main shades with short transitions between them.

I think I'm really impressed by the non-digital twist of the A900 shots I've seen. I'm not a film shooter/fan and I've been known to have use slight oversharpening on my TV set to get a crisp image. But I guess I want the "better than real" level to be just below the threshold of explicit recognition.

P.S.: Hate to point out perceived "flaws" in a favourite image of yours, but just wanted to explain why I personally don't see the IQ in the same league.

[falconeye]

No, I don't believe so.

Incorrect.

DxO is intelligent enough to normalize their DxOMark scale to 8Mpixel:

DxOMark Sensor scale is normalized for a defined printing scenario—8Mpix printed on 8”x12” (20cmx30cm) at 300dpi resolution. Any other normalization, even with higher resolution, would lead to the same ranking, given that any camera which could not deliver the chosen resolution would be eliminated from the comparison.

However, DxO scales results using a theoretical formula rather than scaling images prior to testing.

[falconeye]

P.S.: Hate to point out perceived "flaws" in a favourite image of yours, but just wanted to explain why I personally don't see the IQ in the same league.

Never mind, it is a favourite image of mine independent of IQ

It is difficult to reply to your A900 experience w/o images to look at. Is the one in this thread a good example of what you are talking about?

Normally, the digital vs. analog resemblance is entirely a matter of lighting and post-processing profile, i.e., levels, sharpening and CIE color matrix. There is a possibility that 25MPixel help eliminate the Bayer matrix and pixel artifacts. But this difference should have become invisible at 4 MPixels or less.

[Pentaxor]

by - mviktor

half part of the nose (middle bridge down to the tip) seems to be off-tone and smudgy compared to the exposed side part of the nose showing an obvious outline. transition doesn't look that great on that part. sorry to sound like a perfectionist !

[poppyer]

BTW the ugly table and chart at the beginning of this thread was created by me just as a way of summarising some of the DXO camera results. It was not produced by DXO.

They have a far more sophisticated and useful summary of the low-light ISO camera scores available here:

DxOMark Sensor

no worries, rawr. i like your plot, nice with colors.
the only thing i think is to change to x-axis from linear scale to log scale, i.e., 500-1000-2000, which will avoid some misinterpretation of the numbers.

[Class A]

It is difficult to reply to your A900 experience w/o images to look at. Is the one in this thread a good example of what you are talking about?

Yes, it is a good example. I saw others in a Sony A900 brochure and I think a couple of more on a blog. Someone once did a comparison between the A900 and a Nikon and a Canon, IIRC, and although he came to the conclusion the Canikons were technically the better cameras, he still preferred the Sony for its unique image rendering.

Normally, the digital vs. analog resemblance is entirely a matter of lighting and post-processing profile, i.e., levels, sharpening and CIE color matrix.

It could well be the case that I'm fantasising and that I would mistake a properly processed K-7 image with an A900 shot in a blind test (hey, don't even think about it ). I guess that's even likely but so far the evidence that I saw figures and rhymes with the tester who preferred the Sony despite it losing the technical battle to the Canikons.

But this difference should have become invisible at 4 MPixels or less.

You are right, but what do you think about my "some image properties survive downscaling more than one should expect" theory? It's probably another folly, but again, it seems to be supported empirically.

For instance, I only saw a number of rather small shots of the DA 55-300 on this forum. Still these suggested to me that this lens has way better than average sharpness. Only later I read that this is actually true. Given the sizes of the images posted to the forum, should I be able to tell a 18-250 @ 250 end from a 55-300 @ 250? Probably not, but it seems I could. There could be other explanations for this, of course.

...sorry to sound like a perfectionist !

Nothing wrong with that.

On close inspection it seems that a few local parts look a bit funny but I'm not sure whether that's OOF make-up or some retouching or both. But I don't think that what makes that image sing for me on a global level is the result of PP.

[tigrebleu]

It's weird how the K10D is better than the K20D, and the K20D is better than the K-7. It should be the other way around.

The K20D was a bit more pixel-tight than the K10D, so it's no surprise there, IMHO. Even with larger photosites, like with the Samsung sensor, there are other factors to be taken in account, like the circuit noise-related issues due to heat, external and internal EMG interference, etc.

Color is also more accurate on the K10D, thanks to the CCD sensor, which is much better than the CMOS on that aspect. The CMOS advantage on high ISO noise and power consumption gives it the edge over the CCD, however.

But I love my K-7: it's way better than my K10D in terms of AF and handling, and it's way, way better in low light.

***

I must say I was expecting better results for the K-7 vs. the K20D. But alas, dpreview (and now DxO) showed that the K-7 was a little behind the K20D.

With a 4-channel output, I was expecting lower noise levels at high ISO, thanks to a faster readout, since less heat is generated because of the shorter working time of the sensor + ADC pipeline. But maybe the different circuitry has an influence on the noise: more circuitry is more possibilities for heat to increase noise levels, so a 4-channel vs. 2-channel readout might not always be an advantage.

The K-7 still rocks in terms of features, built quality and general feel and IQ, though.


BTW, the D5000 also shows little or no improvement over the D90, although both cameras share a quite similar sensor. It seems every manufacturer is having a hard time pushing the APS-C sensor beyond its actual performance, and for good reason.

Let's face it, the APS-C has limits. The D300s, 7D and K-7 show pretty similar noise levels in RAW, a sign that we're reaching the limits of physics in terms of actual sensor design.

So unless someone comes up with a new sensor design (the latest EOS 1Dmk4 sensor with gapless microlenses is a good example of a great but nowhere near fantastic improvement) or unless we reduces the number of photosites on actual sensors (not likely to happen soon), the only way we can really improve the noise levels and general image quality is by increasing the size of the sensors. With full frame sensors approximately 2.25x times larger, the gain in noise levels is usually a one or two full stops more than APS-C with a similar pixel count. Figures...

Full frame will eventually become cheaper, and someday FF sensors will be so cheap that their costs will become almost marginal in the whole manufacturing equation. That day, full frame will be able to compete directly with APS-C.

Until then, I suspect all APS-C sensors will be pretty close to one another in terms of image quality (at least when shooting RAW) and that improvements in the departement of noise levels will come more often from in-camera software than from sensor developpement.

[Pentaxor]

Nothing wrong with that.

On close inspection it seems that a few local parts look a bit funny but I'm not sure whether that's OOF make-up or some retouching or both. But I don't think that what makes that image sing for me on a global level is the result of PP.

Thanks! one of the few local parts that kept on getting my attention is that portion just above the left eyebrow (temple area). the color tone seems not to match it's surrounding or uneven shading and like you said it maybe an OOF make-up or some retouching done. it is somewhat obvious because it is kinda soft and lacks some details on that particular portion. honestly, it looked like a bump. I can't say that it's a result of the lighting or just because it is really a bump (which I'm doubtful).

honestly, a great lighting there !

[Jaleel]

In the image you included I don't see the "magic" I see in the A900 image. It is a great shot but it somewhat has written "digital" all over it.

I see the magic only in light and... beuty

[falconeye]

my "some image properties survive downscaling more than one should expect" theory?

You guessed I would jump on this, right?

First and theoretically, an edge which "smears" across 2 pixels can lead to softness even when scaled down to 50%.

To see why, let's look at a possible downscaling scenario:
Sharp edge | 2px edge
100 100 100 100 000 000 000 000 | 100 100 100 050 050 000 000 000
50% -> 100 100 000 000 | 100 075 025 000
Of course, a different placement of the edge would have eliminated the difference, but on average, some difference remains, i.e.:
50% -> 100 075 025 000 | 100 063 037 000
which is a reduction of edge contrast by 50% (50 to 25).

Sharpening can recover some of the original sharpness.

Now, many images smear edges across 4 pixels (20µm, about as large as the circle of confusion for DoF calculations). This should remain visible even in reduction to 25% of original size which is about 800x1200 or web size.

Therefore, your theory survives a first plausability check