Camera site image samples next to worthless (time to standardize?)

It could be time for sites to standardize on a RAW conversion for
image samples because the inevitable JPEGs put up on sites are highly
subjective, unless you only shoot JPEGs. Some camera companies just
can't do good JPEGs. Even a company relatively good at processing
them in-camera (like Olympus) can't always do justice to the imagery
the camera is capable of.
I've seen 200 ISO shots with noticeable noise reduction being applied
(usually, some ill-chosen default setting in the camera) so perhaps it
is time to pick a RAW converter than handles any RAW file and use it
for all camera shots. Inevitably, you'll get people saying you can't
do it, that RAW converters work differently with different cameras,
which is true, but IMO, it would provide a better snapshot of what a
camera can do, and thus would be useful, compared to using JPEG
outputs.

 

Can any JPG do justice to the output of a modern high resolution
camera?

 

If you shoot jpeg as many people do, then it /is/ the output of a modern
high res camera.

 

I can't trust what others do to there JPG images, just to gloss over how
a camera performs.

Using JPGs depends on the final output of the image. I can see no
problem in using my JPG's for printing to a large format image.

Comparing printed images taken in RAW and JPG, processed through PShop I
can see negligible difference. Yes you can tweak a raw more, but its
what's acceptable to the client that matters, not what you put yourself
through because you can see fault.

 

On 23/02/2013 08:22, Eric Stevens wrote:
[]

Yes, as there exists 12-bit, lossless JPEG. But if you mean 8-bit JPEG,
then the answer depends to a a degree on the viewing conditions and
whether there is any post-processing involved. Dynamic range is more
likely to be the problem than spatial resolution.

I use JPEG, and it's fine /most/ of the time.

 

Olympus are con-artists and liars. The stated ISO settings on the
OMD-EM5 are off by more than a stop - ie so-called "ISO 1600" is
actually ISO 782.

So what are you going to compare an "ISO 1600" OMD E raw file with?

Corrected for ISO, a 6 year old and two generations out of date D90
sensor outperforms the OMD (the OMD has slightly higher QE, but more
than offset by a smaller sensor).

 

Ditto for me too.

I used to shoot Raw all the time, expecting the best of what I
photographed; slowly moving to Raw+JPEG, following the advice of
photo-geeks like me. I would process my shots in Lightroom then finish
off in Photoshop. Lots of sometimes tedious work. Then one day I
realized people viewing my prints didn't give a damn what I was saving
the photos with or how I was processing them, they either liked the
scenes/ prints or they didn't. But no one, not ever, (with exception of
the salesmen at the camera store) who liked my scenes/shots asked me if
I was shooting Raw or JPEG. So ... from that time on I wised up and
started using JPEG most of the time.

Facing the issue straight on, when about 98% of those who ask me for
prints (such as friends and family) want their prints made 6x4 size, how
technically accurate does a photo shot have to be. Just get the flesh
colors right and make people look like a movie star and satisfaction
will soon follow. A worked-over scene is either appealing or it's not.
Emotion rules, while in general practice, technology comes in second
place. Have to say, being an armature ofttimes has its good points.

 

I tried to point out mis-statements about ISO with Pentax a few years
back and got attacked for it. Olympus ISO's used be underrated, which
worked in their favour, being the cost of an f/2.8 lens is always 2x
the cost of an f/4.0 lens. Obviously, true ISOs would need to be
determined before doing such a test, otherwise you end up like
Dpreview, trimming up exposures in post-processing to make each scene
look the same, but killing any ability to really compare.

 

Being an "armature"...don't you become far too dizzy, to take steady
pictures?

John

 

LOL ......

There may be an old Chinese proverb that goes something like this:

Without eyeglasses, haste and negligence are companions to one who
enters the land of the fuzzy, ofttimes wearing the crown of laughter.

Bless you, John

 

Well, one good turn deserves another, just don't get too wound up.

 

It isn't the size of the image sensor that limits JPEG performance. It
is the choice of quantisation tables used for compression and a few
minor defects in the usual decoders reconstruction algorithms.

Faults blamed on JPEG are almost always somewhere else in the processing
chain - usually in the demosaic, denoise and sharpen.

The only place where a JPEG really fails against RAW is in images with
huge dynamic range where the shadow detail and highlights are both
important - most obviously in wedding photography. The extra dynamic
range in the RAW data is very useful in those circumstances.

Most camera manufacturers can manage to encode perfectly decent high
quality JPEGs and have done so ever since the first generation of
cameras. Most people would not be able to tell the difference between a
JPEG encoded photographic image and the original reference data.

Such differences as there may be are down to other gratuitous in camera
tweaks like oversharpening and noise reduction tricks. Manufacturers
know that ever increasing numbers of megapixels and sharp edges with
over exaggerated colours sells cameras so that is what they make for the
consumer end of the market. Images taken on pro cameras look slightly
soft by comparison but their resolution is all real.

The main advantage of a JPEG is that it is much smaller for a given
level of image fidelity. You would have to use a *very* much larger
lossless RAW file format for no perceptible difference.

 

That isn't really what I meant. In the early days there was no doubt
that a digital camera was incapable of capturing the fine details of
the image which the lens was throwing onto the sensor. As sensors
improved they became more demanding of lenses but still, the
capability of lenses generally exceeded that of the sensors.

JPG is inherently a lossy system of storing images but in the early
days this didn't matter too much as right down to pixel level there
was not too much fine detail in the image.

Now we have 36MP and 40MP cameras with sensors capable of capturing
every morsel of detail that the lenses are capable of passing on to
them. It seems to me that it impossible for an inherently lossy file
system to record the full range of detail that the best modern cameras
produce. Hence my question:

"Can any JPG do justice to the output of a modern high resolution
camera?"

I guess it partly depends on what you mean by 'doing justice' but it
seems to me to be counter intuitive to build a camera which is capable
of recording all the detail that the best lenses can convey and to
then use an image file system which throws away some of that detail.

 

The JPEG format certainly can, and does, handles a 36MP resolution image
just exactly the same as it handles a 3MP image.

The fact that it is inherenlty lossy has very little to do with megapixel
resolution.

That is true, unless you are claiming that is the case with JPEG. It
is not applicable to JPEG.

 

Err no. You have it wrong. In the early days the pixel size was huge and
the lens was capable of out performing the sensor so that there were
comparatively huge amounts of high frequency content in the raw image.
The anti-aliassing filter was there to make things behave.

Except that JPEG *doesn't* throw away any of that detail. Photographic
image detail *is* preserved by properly executed JPEG encoding.

The first generation losses look for all the world like noise. You would
get bigger differences in the image by waiting 1ms to make the exposure
due to changes in the ambient lighting and subject movement.

The lens MTF is the limiting factor in very high pixel count cameras.
The image from a big sensor is more ideally suited to JPEG encoding than
one taken on a sensor which cannot oversample the image.

Basically you have it completely wrong. Larger pixel count sensors
produce images that have locally correlated pixels which actually
*decreases* the amount of high frequency content in the final image.
This makes JPEG encoding work better on them!

Astronomers spend a lot of time matching their sensors to telescopes to
glean the maximum amount of information about the sky. If you want to
better understand why your intuition is wrong you need to look at the
MTF for diffraction limited optics.

 

It seems to me that 'doing justice' does not include using an
inherently lossy image compression system that entails discarding fine
detail. OK, in some cases it's very fine detail but it very often is
the very fine detail which enables the camera to accurately capture
the texture of surfaces or the details of landscapes etc, among other
things.

It seems strange to me that one should go to the expense entailed in
buying one of the current high-end cameras and then transmit its
images using a file system which effectively reduces its resolution to
that of a much lesser camera.

 

rOn Tue, 26 Feb 2013 08:11:52 +0000, Martin Brown

What you are suggesting is that the fine detail is not present in the
image in the first case. This would be true except in the case of the
very best lenses. But if the lens and camera are equally well matched
there will be no possibility of oversampling and the sensor will
capture all the detail which the lens is capable of conveying. In that
case any loss of information in the image compression process can only
result in degradation of the captured image.

 

Of course, but in fact JPEG does not necessarily
discard more useful fine detail from a high resolution
image than it does from a low resolution image.

That is not what "lossy compression" is about.

As for the difference in how high resolution image are
handled compared to lower resolution images, consider
for a moment the distinction, on a percentage basis,
between using 8x8 pixel blocks on a 3.6MP image as
compared to the same sized blocks with a 36MP image.
Obviously the lower resolution image suffers more than
the higher resolution image!

No it is not.

JPEG is not "a file system". JPEG does not reduce image
resolution more for high resolution image than it does for
low resolution images.

JPEG retains the information that can be used for visual
display of one single image. It does that very
effectively if the "Quality" parameter is set relatively
high, and of course much less so if it is set lower.

You are confusing a RAW data that set contains massive
amounts of data and can produce nearly an infinite
number of different images, and/or a resulting TIFF
formatted image which is indeed only one image but
retains significantly more information than is necessary
to display precisely that image (and thus is a fairly
ideal intermediate format when further editing is to be
done), with the JPEG format which is none of the above
and instead is strictly a final product that is to be
used for display and should not be expected to provide
enough information to allow editing (even though it
commonly does do that too, in a very minimal way).

 

The crucial point you are missing is that lens apertures are round and
as such their MTF tapers off rather rapidly at the highest spatial
frequencies. This is an inevitable result of the physics of diffractive
optics. The best sensors do now oversample the image and even the best
lenses are incapable of reproducing very high spatial frequencies at
high contrast with anything close to full amplitude.

Whilst I don't agree with everything they say this site has a reasonable
introduction to lens MTF graphs and a few examples of the spatial
frequency transmission.

http://www.cambridgeincolour.com/tutorials/lens-quality-mtf-resolution.htm

The important thing to remember is that the lens is not capable of
conveying very much high frequency signal to the sensor. This makes the
JPEG encoding much more effective on large sensors in practice.

But don't take my word for it. Do the experiment with a test chart. I
think you will be *VERY* surprised how little high frequency content
there is in an image made with a high MP sensor.

You will also find that handheld without a tripod it is even more
difficult to record all the high frequencies.

You are tilting at windmills with a very limited understanding of what
you are talking about. The JPEG image does not throw away fine detail
and to get the optimum match of optics to camera you must have at least
1.5x oversampling - scientific research often uses 2x to make sure that
post processing can take advantage of deconvolution codes.

The lossless full dynamic range image may allow a lot more post
processing to bring out shadow detail without risk of seeing JPEG
artefacts when you contrast stretch but that is not to say that the JPEG
has lost any fine detail. Mostly it has lost noise which in a
photographic image helps dither the apparent colour and prevent visible
contouring. Most people cannot detect a JPEG image in a blind test.
If you really want to understand this stuff you need to look at how the
astronomers match their essentially perfect optics that are limited only
by the physics of a finite aperture to a large CCD sensor.

Camera lenses are crude compromises by comparison as there is always a
penalty incurred to get a large flat focal plane and no vignetting. When
the lens is stopped down at some point the geometric optics ray tracing
aberrations and diffraction spot become roughly equal and that is the
sweet spot for the lens. Most very high quality lenses this is usually
around f5 to f8 but there are a few exceptions.

 

How many more times do you need telling? JPEG does not discard fine
detail - it works in the frequency domain and exploits the limitations
of the human eye to obtain maximum fidelity in minimum size. There is no
point storing fake detail arising from thermal noise in the sensor.

The main hit in storing as JPEG is the dynamic range gamma correction
which if done badly in camera will lose one or other of highlight or
shadow detail irreversibly. It is not rocket science to spot the
situations where this is likely to be a problem. Bride in white or cream
gown and groom in black velvet suit is the most common example.

Except that it doesn't. The JPEG encoder works better on larger sensors.
Lens performance is well down at the highest spatial frequencies - even
for the very best prime lenses.