PanoTools:
Re: My attempt to eliminate vignetting effect
JD Smith 2004-Dec-01 06:17:47
On Tue, 30 Nov 2004 11:54:25 -0700, Erik Krause wrote:
> On 30 Nov 2004 at 8:26, JD Smith wrote:
>
>> In reality, most people won't want to do this, so some in situ
>> estimation of a rough radial vignetting correction function may still
>> be appropriate. For typical 30% overlap, the exact center of the
>> function probably won't matter too much.
>
> What do you think: In a 30% overlap is there enough information to do a
> fairly exact determination of vignetting? I think it should be possible
> but that's only a feeling...
Well, it depends on what you mean by exact. If you come much closer
to the structure of the mismatch by assuming a radially symmetric
function centered at [w/2,h/2], you should be able to make real
improvements to the results enblend delivers, which are already very
good. Remember that enblend knows nothing about the large scale
spatial structure of the mismatch.
On the other hand, the eye really doesn't care about smooth rolling
variations in, e.g. sky level over 10s of degrees, which is why
enblend already works as well as it does. If you really want this
correction to remove exposure variations in panoramas obtained without
fixed exposure value, you do run into an issue that HDR users are
confronted by: how to represent the dynamic range present in a
vignetting+exposure-corrected image, which now exceeds what's
available in typical 8bit output (or displays, for that matter). In
other words, in order to avoid clipping highlights and low-lights
without resorting to fancy HDR techniques, it may actually be better
just to leave adjacent images with exposure differences (say a factor
of 2) as is, and just blend them as best as possible. I suppose you
could fit the a_0 scaling parameter in my example, but not apply it,
but that still leaves a very large "step" for enblend to smooth over.