Multipoint AF tricks?

[Lynn Towns]

You stated this very well, Mike. I'm in total agreement.

Lynn

 

[Mike Fitzgerald]

My understanding of the contrast based AF is that the camera tries
to maximize contrast in the area of interest. At maximum contrast
there should be focus.

That's pretty well it. Definitely (and obviously enough) with the "contrast detection" AF systems that have been the norm with consumer digital cameras; but contrasting material also counts with phase detection systems in dSLRs that still need clearly delineated elements in the subject in order to work well.

I believe some cameras limit the focus
range in different modes, for example a portrait mode may limit
focus from 2 to 10 feet to avoid locking on to the background.

Yes, I've seen reports of moves in that direction, and such would be a definite plus in whatever real advantages one might find in a meaningful set of "Scene" presets. It would make a lot of sense in a Portrait preset.

This is getting into the more sophisticated end of multi-zone AF exploit, however. A single zone (if it's truly a single zone and not subdivided) can't readily be used within this constraint, since it would cause the AF system to want to hunt (and therefore be potentially slower) when the zone is partially off target and can also see a more contrasting background.

I imagine if the camera looked at all focus distances for contrast,
there would be a maximum for each object when it came into focus.
For example if you had an object at 5 feet and one at 30 feet there
would be a bump up in contrast at both distances. If the camera
had all this information it could choose the 5feet object as
nearest. With the current system it chooses the brighter contrast
between the two objects.

The gimmick is that the camera needs to determine the actual distance for each of the objects causing those bumps in contrast, and they may still be swamped by an unusually busy or contrasty background. With a traditional contrast detection AF system the drawback is speed, since the only way that the camera can dermine distance for each of those objects is to adujst focus until it (specifically) achieves peak contrast (we're assuming that each of them is represented in a focus zone) and then read back the distance from the lens assembly. The system would have to move through each object in turn until the distances for all of them have been logged. It can be programmed to reject any that fall outside the range, but it may still need to assess all (or nearly all) of them to make a useful decision. A potentially slow business.

This is my issue with the 707. The camera locks onto a bright
background even if it only covers 5% of the focus area. I usually
avoid this by zooming in and out or choosing a large target but
when I miss, I might not see the issue until I upload to a large
computer screen. Zooming in and out also takes time and requires
independant steps for focus and exposure metering.

A sound understanding of you camera's idiosyncrasies is invaluable. I think the most important single factor to pin down empirically in an AF system is the true working area of your marked focus zones. In some cameras they are very sharply and accurately defined by the AF rangefinder brackets (my previous Coolpix 5700 was a good example of precision in this department). In others you'll find that the effective target area extends well beyond the brackets or other markers.

Knowing just how far it extends is the key, IMO, to quick and predictable use of any AF system. And the same goes, of course, for spot metering systems.

I've never done any detailed experiment with my 10D, but its quite obvious that the camera is sensitive to things beyond whichever of the little AF squares you have active. At least I can see what's happening in the excellent viewfinder that the 10D has, and re-aim slightly to skirt the confusing region; but it's been a regular topic of heated user discussion.

--
Mike
Melbourne