At 2/4/2025 06:19 PM, an OM friend wrote:
>> What MikeG is saying is that it does make a difference, blurring the
>corners, in his experience.
>
>That absolutely used to be the case. I've used Lightroom almost exclusively
>for the past seven years and have found that this algorithm and the lens
>profiles has gotten so good that the corner degradation is largely a thing
>of the past. Not totally, but with these latest/greatest (or semi-greatest)
>lenses from Sony, the results are excellent. That said, my 28-70 kit lens
>totally sucks pond water. Well, it sucked ocean water...
>
>
>> Digital rejiggering of images decreases sharpness. That's just the way it
>is. MikeG can do the math; me, not so much.
>> Ctein contends that Bayer array color loses about as much fine detail as
>halving resolution, because of the interpolation.
>
>So, interesting thing about that. I called Ctein out on it and he
>acknowledged that it was based on one style of algorithm, but then defended
>it by saying "well, the color information still has lower resolution."
>
>It depends on which converter, and which version, and which camera sensor.
>So nobody is right, and nobody is wrong.
>
>Converters can use a 3-pixel merge or a 4-pixel merge. There is such a huge
>misconception about the 2x green pixels and everybody thinks that it has
>something magical to do with the luminance information used by the green
>channel and so forth. Actually, the reason for the 2x green pixels is that
>when we use the 'nearest neighbors of the "intersection" of the pixels,
>there is always one green, one red, one blue pixel to work from. The
>lumenance and color is derived from the averaging of those three pixels.
>Not four. It's a rare converter that actually uses 2 green. One of the
>problems with the Olympus E1 was that it NEEDED a 4-pixel merge in order to
>keep from turning the picture into a mosiac. Converters used to treat
>luminance separately from color, but they are now mostly done together and
>is based on the three-pixel merge for both.
>
>As to the 50% loss of resolution, I would disagree because the empirical
>evidence shows that with the application of the appropriate sharpening at
>the pixel level, we can achieve per pixel edge definition. The difference
>between what Oly's HR mode and a standard merge is doing is that HR mode
>does not require the corrective sharpening. That said, where HR mode really
>shines is the increased depth of color and glassiness to the image which is
>likely a result of the oversampling that is occuring.
>
>Which brings me to the parallel. I'm deep deep deep into music production.
>One of the things that completely transformed digital audio was
>"oversampling." 30 some years ago I helped create one of the first
>oversampling algorithms. (A different better one became the standard, oh
>well). Anyway, the general ideal behind oversampling is to better represent
>what is going on BETWEEN the samples. Without oversampling, it was the job
>of the low-pass filter to eliminate all one-sample transients as well as to
>prevent aliasing. This low-pass filter effectively reduced the TRANSIENT
>frequency response of the audio signal by a minimum of 50%, and a more
>typical maximum of 90%. In other words, a 44.1kHz song on a CD would have a
>sine-wave frequency response of around 20Hz to 20kHz, but the transient
>frequency response (square wave, saw wave), was 2kHz or lower. This not
>only affected the type of sound being played back, but also the position of
>the sound. Phase coherency was lost. Through oversampling and the
>elimination of hardware-based low-pass filters, we are able to not only
>maintain more phase coherency, but the position of the sound is maintained.
>I'm oversimplifying to the point of inaccuracy, but you already know that's
>what I tend to do.
I'm a sony convert so I feel OK to jump in...
I once was an engineer, designing over sampled A/D converters (switched
capacitor analog circuits). The great thing about over sampled A/D is the
anti-alias filter in front could be very simple. Image sampling is the same
only in two dimensions. Hence the anti-alias filter on camera sensors.
One thing most don't realize about filters is the transient response, which
you alluded to. With a non-oversampled A/D, the anti-alias filter needs to be
sharper. However, the shaper a filter is the more over shoot there is in the
step response. If you restrict the step response of a low pass filter to no
overshoot, it smooths out the frequency response with the ideal being
a bessel filter. Not a very sharp filter frequency wise, but a very pleasant
transient response.
I'm sure the sound of sharp filters on transient sound is noticeable. It is
similar
to over sharpening an image. Perhap Heisenberg has something to say about it.
I notice You are using an A9, which is 24mp. A7Rv is 62mp or about
1.6 times more linear sampling. Sony has to make sharp lenses if
they are to match their cameras. My go to lenses are the 20-70/4 and
the 70-200/2.8 II with sometimes adding the 1.4x. Everyone has
there preferred system and it is a wonderful time for photography.
When I go back to my A7iii images, 24mp, I rarely feel there is not
enough resolution.
>I bring this up because with Lightroom's latest/greatest converter,
>especially with the SONY cameras, it is actually performing a multi-pass
>process which does the normal convert of the nearest neighbors, but then
>does a second convert to calculate the in-between values. This is then
>combined to find a better value for the resulting pixel. I haven't figured
>out how much it is applying, but it's definitely throwing extra spices into
>the pot of chili.
>
>So, related content: Yesterday, I met Wayne and his lovely wife at their
>home north of Melbourne. While no pixels were harmed, we had a terrific
>lunch as we realized that this Olympus list friendship has gone on for
>almost 30 years! I had to cut our time short because I had committed to
>helping my daughter move apartments (why oh why did I do this to myself?).
>This morning, I'm sitting in a cafe working on a song that will be released
>later this month.
>
>OZ Schnozz
>--
WayneS
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|