Is CDAF really derived from in-cam JPG adjustments?

Started May 27, 2012 | Discussions thread
Flat view
Detail Man
Detail Man Forum Pro • Posts: 16,683
Is CDAF really derived from in-cam JPG adjustments?

This is not to doubt or question those who are finding that an Olympus "Vivid" in-camera JPG mode seems to improve C-AF results (or even S-AF results). That's fine, and I do mean to "rain on your parades" (so to speak). I presume that such assumtions are based upon adequate experimental "controls" (that is, trying the very same thing when not in the "Vivid" mode, etc.) ...

This is about the general implications regarding the Auto Focus "system architecture" of various brands and camera models - because it clearly has important implications for the performance characteristics of CD Auto Focus systems in general (as it affects not only JPG, but also RAW, recording). The typical speed of CDAF with many M43 systems makes these things hard to determine via time measurements - but the issue of auto-focus integrity remains paramount indeed

It seems reasonable to assume that some (possibly abbreviated for speed) form of de-mosaicing of the photo-site data occurs prior to evaluation by a CDAF system implemented in a camera system.

We know that most/all systems widen the aperture to it's maximum diameter when auto-focusing (and at least in some cases when manual focusing). This makes good sense. Maximizing the absolute value of the (de-mosaiced) pixel-data (assuming that photo-site saturation does not occur as a result, that is) should result in better performance of the image-data utilized in CDAF systems.

It makes sense that in-camera JPG adjustments that directly and most significantly affect the amplitude of the image-data anywhere in the signal pathway preceding the CDAF system would/should be taken into account (so as to avoid non-linear "clipping" at any point). I'll buy that.

But, what about some of the other in-camera parameters that are user-adjustable ? " Contrast " ? In my Lumix FZ30, FZ50, and GH2, the "Contrast" adjustment appears to modify the scaling of the tone-curve below the maximum point of the tone-curve transfer-function - but does not alter the set-point of the maximum point itself (so as not cause large changes in the overal "brightness"). Nevertheless, it's hard to understand why camera designers would want "Contrast" adjustments to significantly affect the CDAF system (causing variations in CDAF performance based upon settings).

" Saturation " ? It seems unclear whether the CDAF should (or does) care about relative intensity of the color data. CDAF would rightly seem to concern itself with composite luminance - and allowing color saturation adjustments to alter CDAF functionality would (again) only seem to cause unwanted variations in CDAF performance based upon settings. Again, this does not sound wise ...

" Sharpness " ? Here again we have a situation (probably the most dramatic case of all discussed here) where the operational integrity of the CDAF system would become hostage to user-settings.

It makes zero sense to me that camera designers would prefer a situation where users become forced to use specific "modes" (such as Pana "Film Modes", or the Oly equivalent seen in modes such as "Vivid") in order to try to coax better (or even adequate ) CDAF system performance out of their cameras. Most users would miss such subtleties altogether - and simply conclude that a camera model has "lousy AF". A tiny, tiny fraction of users who lurk and post in fora such as this would endlessly trade in "groovy" in-camera JPG user-settings that might allow them to actually achieve varying ( possibly better) levels of CDAF system performance. A subjective "total mess" ...

Nothing have to do with common sense (engineering, or otherwise) would imply that such a cloudy and variable situation where it comes to CDAF system performance would be a desirable or a prudent one where it comes to camera design and utilization. RAW shooters (who thought that they had blissfully avoided all the in-camera JPG settings "stuff" and worries) would be forced to join the fray of confused and non-scientific speculation as well - simply to try to ensure that their cameras' CDAF systems would/will function in an effective and acceptable manner . It is (sort of, in some cases) well understood that in-camera JPG settings affect previews/histograms/blinkies, and any other matter surrounding camera metering, (and the RAW-level scaling) - hence, "UniWB" ...

But, for camera designers to extend such a bona-fide "total mess" (surrounding in-camera JPG settings) beyond (camera-setting determined) "exposure values", and into matters of the operational characteristics/integrity of CDAF sounds like a prescription for trouble/disappointment ...

A few posters have in the past boldy claimed that because "preview" modes have to operate quickly , then it would follow that it would seem to be unlikely that the image-data that CDAF systems operate upon could itself be anything other than what the user (and histograms/blinkies) "see" in LCD/EVF displays. Yet, we are talking about proprietary system architectures that are very closely held by camera manufacturers (of which we can only merely speculate, and try to somehow evaluate by some sort of observation of how the systems as a whole operate). In one case, a poster who merely speculates that the above-described situation is likely the case also commonly asserts that the chip-sets in camera hardware-processors are very complex, capable, able to simultaneously perform a number of operational tasks simultaneously . Thus, much remains unclear.

So, a number of things about these matters simply "do not add up" - on an engineering design level, or on a practical user level. "Experimental observations" conducted which do not implement adequate "controls" (images also recorded without such "special settings") seemed doomed to the realms of potential (and signifcant) subjectivity . Can "techno-wizards in the know" enlighten us ???

DM ...

Panasonic Lumix DMC-FZ50 Panasonic Lumix DMC-GH2
If you believe there are incorrect tags, please send us this post using our feedback form.
Flat view
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow