Re: [gentoo-user] merging or fitting images together

[Michael Mol <mikemol@gmail.com>]

On Sat, May 19, 2012 at 10:01 PM, Stroller
<stroller@stellar.eclipse.co.uk> wrote:
>
> On 19 May 2012, at 20:28, Michael Mol wrote:
>> …
>> Worse, if the photographer was not using a prime lens[1], and was
>> instead using a lens with variable zoom, you can't easily know what
>> the real focal length was, as this will change depending on how far
>> the photographer has zoomed in.
>
> Throughout everything else you said I was thinking something like this.
>
> Zoom lenses were much less common even 2 or 3 decades ago.

Ah. Excellent point!

>
> For a long time, a 50mm prime was the common kit lens, rather than the 18-105mm zoom which is sold today.

18-105? I'm used to seeing 18-55.

>
> This was because, on a camera using 35mm film, a 50mm focal length gives a field of view very close to that seen naturally by the human eye.
>
> Wikipedia states that "the first modern film zoom lens was designed around 1950 by Roger Cuvillier" and Canon's official website (the "Canon Camera Museum" pages) states that "The history of Canon's zoom lens goes back to 1954."
>
> Since the photos are stated go have been taken in 1953 it seems highly unlikely that the photographer was using a highly expensive and cutting-edge zoom lens. I doubt many people would have been able to afford these zoom lenses when they were first released.
>
> It seems to me safer to assume that the lens is a 50mm.

Probably generally true. (Though as Philip later remarked, it turns
out the lens was likely a 75mm prime)

>
> I guess focal length may change fractionally during focussing (as lenses are moved back and forth during as the focus ring is turned), however it may also be that a camera manufacturer designs a lens with a 48mm focal length (because that's easier to construct for some reason, or produces better images) and decides to sell it as 50mm because a 2mm difference in focal length makes no difference to the photographer.
>
> Or it may be that the distortion is caused by lens distortion - perhaps Hugin is trying to compensate for that, and straightening up lines.

I don't think the 'Align' button in the wizard tries to optimize for
lens distortion...adjusting for lens distortion tends to take a fair
amount of time in terms of CPU, and far longer in terms of finding the
right sequence of control point optimizers, where an errant point
won't send the algorithm into mathematically weird territories.

>
> In any case, I might try re-doing the stitch a few times, each time telling Hugin the lens is 47mm, 48mm, 49mm, … 51mm, … 53mm. Perhaps you may find that one of those is perfectly spot on.

I tried it again, this time using 342 control points generated by
hugin-cpfinder and autopano-sift-c. (Several runs of the latter, with
100 points each, produces a good set of points). There wasn't anything
for celeste to pick  up, so I used the "fine-tune all points" tool,
and then cleared the thirteen or so points which didn't have good
correlation.

Following that, I ran the control point optimizer in "anchored,
positional mode", checked the preview, and then ran the "everything
without translation" control point optimizer. Checking the preview
again, the panorama was way off-center, so I dragged it back into
place using the fast preview window.

Following that, I ran the exposure optimizer's "low dynamic range"
preset. In the preview, things looked OK. The leftmost portion will
never look all that great, as he captured the sun setting behind a
building (or maybe that's a water spot); that narrowed the usable area
of the dynamic range of the frame, and it's going to look kinda
grayish. If these source JPG files are scans of paper photos, I could
do a lot more with a new scan set using 16bpp TIFF or OpenEXR, and at
perhaps 600 or 1200 dpi instead of 72 or so. Might be able to recover
more detail out out of that leftmost section.

Anyway, the final Hugin pto file is here: http://pastebin.com/gudxvAEa

And the final stitch is here:
http://img407.imageshack.us/img407/2030/brum3068brum30702.jpg

Interesting exercise! The image is still a bit smaller than my first
pass (704x407 vs 785x413), but it's not cropped as tightly, the lines
on the tram are much straighter, and most of the nasty noise on the
leftmost portion has been dealt with. There's likely something that
can be done to blow the image up a bit more.

-- 
:wq