[Elphel-support] latest blog post

Andrey Filippov andrey at elphel.com
Thu Nov 18 09:50:43 PST 2010


On Thu, Nov 18, 2010 at 9:56 AM, Sebastian Pichelhofer <
sebastian.pichelhofer at gmail.com> wrote:

> I see you are just finishing the layout of the new post!
>
> May I suggest to put a full resolution sample (thumbnail, full size
> following link) image "before" and "after" at the very top as well?
>
> Regards Sebastian
>


Sebastian,

there are couple problems with that. First, the standard VNG used different
processing chain (you manually had to to adjust colors, so for the animated
gif I had to use comparison with my (to completely finished) de-bayer, not
the VNG. With VNG the colors "jumps" distract watching the differences:

Here are those files, yo can see for yourself:

http://blogs.elphel.com/wp-content/uploads/2010/11/processing-comparison-vng.gifcomparison
of the final image with the VNG
http://blogs.elphel.com/wp-content/uploads/2010/11/processing-comparison-scissors.gif-
comparison of the final image with just the de-bayered (with
"scissors")


Another problem for the full size/full resolution images is that in the
processing chain I used, the resolution is doubled at the very first stage,
de-mosaic and deconvolution - they all are performed on the 20MPix images.
With the reference VNG we have 5 MPix one. What would be the best way to
compare - up-scale VNG to 20 MPix or down-scale deconvolved ones.

And the last one - the images I used are not so good quality overall, the
target distance is wrong to say nothing that they have a lot of junk
scattered in out officie. We just ordered fancy prints of the larger test
patterns so this weekend we plan to calibrate the actual Eyesis camera at 6m
distance (combining multiple tiles as the pattern is just 2.6m x3.0m) - that
would allow us to apply correction to the actual footage with some nice
outdoor view.

Andrey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://support.elphel.com/pipermail/support-list_support.elphel.com/attachments/20101118/900ba725/attachment-0002.html>


More information about the Support-list mailing list