High DR is just the trick

Started Nov 8, 2010 | Discussions
Oleg_V Contributing Member • Posts: 627
High DR is just the trick

Sorry for not sharing all that hype about fascinating DR results of the K-5 but its nothing more than simple mathematical trick. It was first published by me at the Russian pentax forum almost a year ago http://www.penta-club.ru/forum/index.php?showtopic=69565 . I'm wandering if sony engineers are reading this forum or not since the trick is now implemented in the nikon d5000, pentax k-5 and probably nikon d7000.

The DR definition.

See http://dxomark.com/index.php/en/Learn-more/Understanding-DxOMark-Database/Measurements/Noise

DXO measures the saturation light level L0 and the light level L1 for which S/N=1. Their ratio L1/L0 corrected for the pixel count is DR rating by dxo version. So, whats wrong with them? Mixing the light levels with plain sensor output is prone to tricks. We just consider the simplest of them.

The trick.

Lets add some constant offset S0 to the output signal S = S + S0. In case S0 > = N the DR measured by dxo procedure just goes to infinity. Varying offset one can obtain just arbitrary results. I guess sony is using some adaptive algorithm to obtain close to linear DR curve. The offset seems to be hard limited since it obviously breaks colors so that it becomes smaller than dispersion and so insignificant at high ISO. As a result the DR curve is changing from classical saturated at low ISO form to almost linear with slight kick up at base ISO.

The proof.

The proof comes from Nikon forum recently http://forums.dpreview.com/forums/readflat.asp?forum=1034&thread=36853749 . The d5000 owner demonstrated carefully shot sequence of heavily underexposed shots that clearly shows adding constant offset to the dark level which exhibit itself by gradually reducing image contrast without dropping dark areas to the real dark (0,0,0). The following picture comparing 13 and 14 EV corrected shots just speak for itself.

Why its bad.

The trick just disturbs colors without adding anything useful to the image. Moreover it gives some marketing advantage to those who sacrifice image quality for stupid tests with inadequate method.

laurynas Regular Member • Posts: 287
Re: High DR is just the trick

can you recover 9,5EV from D5000 like we just had en example shown here?

I do not care what they do, but i love all this shadow detail that can be easily recovered

Ivan Glisin
Ivan Glisin Contributing Member • Posts: 726
Re: High DR is just the trick

Very interesting post, but the very end left me a bit puzzled:

Oleg_V wrote:

The trick just disturbs colors without adding anything useful to the image. Moreover it gives some marketing advantage to those who sacrifice image quality for stupid tests with inadequate method.

If I am reading this well, what you are saying is that K-5 could have been engineered to produce even better image quality , just if Pentax has decided not to apply the trick mentioned?

But all examples I am seeing do look really great, so I am curios to learn where an improvement would have been if no trick applied, and what benefit it would bring to image quality? Lower noise, higher DR, better color accuracy, improved resolving power? Perhaps improvement could be added through a firmware update later on when the dust settles? Could you please elaborate?

-- hide signature --
awaldram
awaldram Forum Pro • Posts: 13,271
Re: High DR is just the trick

An interesting theory, a bit ambitious to state it as fact though.!

Especially when the images from the cameras back up the DR figures obtained.

How does your trick translate to great DR with fantastic colour sensitivity and tonal balance.?

 awaldram's gear list:awaldram's gear list
Pentax Q Pentax K-3 Pentax Q7 Pentax K-x Olympus PEN E-PM2 +17 more
gillbod Senior Member • Posts: 1,598
Re: High DR is just the trick

can i just check that i understand what you're getting at? basically, if a constant signal was added to a sensor's output, then it would look like the signal-to-noise ratio was better than it is, since the constant additional output would look like signal. as a result, you might think that you are getting a good signal-to-noise ratio when in fact you are just looking at a bit of gray added in by the camera maker?

but, i don't see any obvious evidence of a constant signal added to those pictures. that you show.

also, what's this variable N that you bring in?

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

gillbod wrote:

as a result, you might think that you are getting a good signal-to-noise ratio when in fact you are just looking at a bit of gray added in by the camera maker?

Exactly.

but, i don't see any obvious evidence of a constant signal added to those pictures. that you show.

The decreasing contrast is a pure artifact. In case the contrast curve is linear the contrast should not depend on the exposure correction. The dark is expected to be dark. Instead it become the more and more lighter.

also, what's this variable N that you bring in?

Noise.

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

Ivan Glisin wrote:

If I am reading this well, what you are saying is that K-5 could have been engineered to produce even better image quality , just if Pentax has decided not to apply the trick mentioned?

Sure. Don't adding artificial dark offset retains colors in the dark areas but gets lower dxo DR rating.

Atindra Veteran Member • Posts: 7,408
Re: High DR is just the trick

Firstly I dont see any Image Quality sacrifice for 14 EV DR in K-5 so far from the images posted on forums, flickr and elsewhere.

Second, everything is a trick. Whole Dslr design is a trickery so what is wrong in this particular trick?

Forget about numbers, forget cameras just see pictures in various challenging conditions, both K-5 and D7k do show remarkable highlight headroom, if they do this by a trick who cares?

If you plan a new measurement that gives DR of 9 or less to K-5, still image quality speaks for itself.

I am happy that Sony, Nikon and Pentax has found out to fool DxO labs.

Atindra

 Atindra's gear list:Atindra's gear list
Pentax K-70 Pentax 55-300mm F4.5-6.3 ED PLM WR RE Pentax smc DA 18-135mm F3.5-5.6ED AL [IF] DC WR
Ron Kruger Senior Member • Posts: 1,975
Re: High DR is just the trick

Thanks Oleg. This explains the difference I have been seeing in the colors. I fear that in the race to compete with the high ISO craze, they have been sacrificing the Pentax look. The difference I notice in the K5 is not as dramatic as with the Kx, probably because of increased megapixels?

Nevertheless, K5 images look pretty good and not quite as plastic as Nikon or Canon, so I plan to get one and figure I can tweak the in-camera settings and PP to get a rich look.
What do you think?
--

In the end, the only things that matter are the people we help and the people we hurt. http://pa.photoshelter.com/user/ronkruger

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

Atindra wrote:

I am happy that Sony, Nikon and Pentax has found out to fool DxO labs.

Sure thats funny, but at the expense of the shadow colors.

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

Ron Kruger wrote:

What do you think?

I think pentax made the good camera, much better than predecessors. Definitely good choice.

gillbod Senior Member • Posts: 1,598
Re: High DR is just the trick

Oleg_V wrote:

The decreasing contrast is a pure artifact. In case the contrast curve is linear the contrast should not depend on the exposure correction. The dark is expected to be dark. Instead it become the more and more lighter.

i'm not sure i agree. shouldn't the contrast depend on the exposure correction? i.e. larger exposure correction leads to more noisewhich leads to less contrast. i don't know how the results you show necessarily mean that a constant amount of grey was added to the output.

also, shouldn't this also mean that it is impossible to register black on the output? i'm pretty sure every camera i have ever used could register black in the output.

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

gillbod wrote:

Oleg_V wrote:

The decreasing contrast is a pure artifact. In case the contrast curve is linear the contrast should not depend on the exposure correction. The dark is expected to be dark. Instead it become the more and more lighter.

i'm not sure i agree. shouldn't the contrast depend on the exposure correction? i.e. larger exposure correction leads to more noisewhich leads to less contrast. i don't know how the results you show necessarily mean that a constant amount of grey was added to the output.

The negative EC was compensated by post processing. So provided that contrast curve is not tweaked the images should be identical except for noise that you can just eliminate by averaging over some image area and compare the mean signal value. Increasing the dark area average luminance is clear indication of the positive black level offset applied.

also, shouldn't this also mean that it is impossible to register black on the output? i'm pretty sure every camera i have ever used could register black in the output.

Sure black is more or less black till you don't push it. Pushing may disclose many interesting things. K-20 used to just kill shadows applying negative black offset to hide shadow noise. But such approach does not improve DR rating. Now the trick is different - just adding positive black offset improves score dramatically. The shadows are preserved but colors are not. So I'm wandering if they give us live signal someday or not.

unlimited New Member • Posts: 6
Re: High DR is just the trick

Oleg,

Can we see that S0 constant offset in the data taken from black (unexposed) borders of raw files (as I remember that was the source of Gordon'sBGood conclusions)? Or it is some sort of adoptive algorithm which is used to add that offset?

Regards
xdrew

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

unlimited wrote:

Can we see that S0 constant offset in the data taken from black (unexposed) borders of raw files (as I remember that was the source of Gordon'sBGood conclusions)? Or it is some sort of adoptive algorithm which is used to add that offset?

If its a trick the offset is the result of the adoptive algorithm. I'm not big expert on image border. We'd better wait for the Gordon'sBGood opinion. The only problem I see - the border may be tweaked as well ..

gillbod Senior Member • Posts: 1,598
Re: High DR is just the trick

Oleg_V wrote:

The negative EC was compensated by post processing. So provided that contrast curve is not tweaked the images should be identical except for noise that you can just eliminate by averaging over some image area and compare the mean signal value. Increasing the dark area average luminance is clear indication of the positive black level offset applied.

the increase in dark area average luminance can also be down to noise. that's what i was getting at. more noise moves the mean luminance of the shadows towards the mean luminance of the noise.

simple experiment. compare the histograms of

1. a black frame in photoshop

to

2. a black frame in photoshop with lots of noise added. the effect of the noise on the mean luminance in shadow areas should then be obvious.

i won't say there is no black offset at akk. i will just say that i feel there are other explanations for what we're seeing in the images you have in the op.

geoffire Regular Member • Posts: 158
Re: High DR is just the trick

Are you suggesting the Sony A580 version of this sensor should actually be better than the D7000 and K-5 versions in usable shadow detail?

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

gillbod wrote:

the increase in dark area average luminance can also be down to noise. that's what i was getting at. more noise moves the mean luminance of the shadows towards the mean luminance of the noise.

By the common definition adopted by dxo as well the mean luminance of the noise is exactly zero. On the other hand provided that the noise dispersion is greater than the mean signal some values may be clipped at the zero level. But on the above image I don't see much clipped points at all.

OP Oleg_V Contributing Member • Posts: 627
Re: High DR is just the trick

geoffire wrote:

Are you suggesting the Sony A580 version of this sensor should actually be better than the D7000 and K-5 versions in usable shadow detail?

Probably yes.

Iliah Borg Forum Pro • Posts: 27,580
Re: High DR is just the trick

geoffire wrote:

Are you suggesting the Sony A580 version of this sensor should actually be better than the D7000 and K-5 versions in usable shadow detail?

It is also better in colour reproduction across the range of densities; but is about 1 stop useful photographic dynamic range less.

-- hide signature --
Keyboard shortcuts:
FForum MMy threads