[Elphel-support] Questions about JP4

[Andrey Filippov]

On Tue, Nov 29, 2011 at 10:44 AM, Florent Thiery
<florent.thiery at ubicast.eu>wrote:

> Hello
>
>    - is a (jpeg) compression applied to the images when grabbing a JP4
>    stream ? In other words, is the compression rate parameter having an effect
>    in JP4 mode ?
>
> Florent, yes, JPEG compression is applied and the quality parameters is
applicable. This quality is calculated according to JPEG suggested
formulas. It is more tricky but still possible to use arbitrary
quantization tables. When you specify quality 100% there is no
quantization, but the files will be large.


>    - (how) can i render a JP4 image to RGB using ImageJ plugins ? I can
>    successfully import it (but it is still gray with squares)
>
> In ImageJ you are getting just raw Bayer mosaic data - same as the sensor
outputs. That is the whole goal of the JP4 format - preserve as much as
possible of the actual senor data while providing reasonable
compression(determined by your bandwidth budget). Normal JPEG is designed
primarily for image distribution, JP4 modification is camera-centric,
providing a way to get data from the sensor to the post-processing computer.

Converting from Bayer mosaic is not a trivial task, there is no single
"best" conversion - there are many different de-mosaic or de-bayer
algorithms, each with different features. Some may work better for nature
photos, some - for photos with multiple straight and parallel lines (such
as text, or photos of man-made structures, like buildings or fences). I was
able to find some ImageJ de-bayer algorithms, but they were rather basic.

When processing Eyesis images each color component is first processed for
aberration correction separately as there is always some of chromatic
aberration in the images. At the next stage the colors are combine with the
frequency-domain algorithm we called "spectral scissors" - aliases are
removed according to the nature of the image, and the shape of the
separating "cuts" for red and blue components (they have less resolution
than green and so closer/more overlapping aliases) are calculated using the
information from the green channel. This implementation is designed to work
on subdivided pixel grid - during the aberration correction we convert each
color component to 20MPix images (2x in each direction) and then reduce
resolution to the original only on the very last stage of the processing.

You may find illustration of this de-mosaic process in the blog post -
http://blog.elphel.com/2010/11/zoom-in-now-enhance/ (around last 1/3 of it)
, the image itself is here -
http://blogs.elphel.com/wp-content/uploads/2010/11/scissors.png  It is
possible to run this algorithm on JP4 images without aberration correction
with
http://elphel.git.sourceforge.net/git/gitweb.cgi?p=elphel/ImageJ-Elphel;a=blob(correction
requires kernels that are calculated with
Aberration_Calibration.java




>
>    - is there a C implementation somewhere similar to our JP46 to bayer
>    code<http://code.google.com/p/gst-plugins-elphel/source/browse/trunk/jp462bayer/src/gstjp462bayer.c#856> so
>    that we can port it ? If not, where can we have more information regarding
>    the difference between JP46 and JP4, and the according decoding adjustments
>    needed ?
>
>
Yes, there is a C implementation of the JP4 conversion, but I would
recommend you to keep an eye on the java implementation in ImageJ plugin.
Once in a while we add more parameters that are passed from the camera via
Exif MakerNote  field, and the ImageJ implementation is always maintained
to match the latest camera firmware. And it still is compatible with the
older firmware releases - so far we were appending new data to the end of
the MakerNote, so the beginning of it stays the same.

You may find C-code here:
http://wiki.elphel.com/index.php?title=Movie2dng(or just search for
movie2dng ) , but it may be a little behind the latest
MakerNote.

Andrey

>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://support.elphel.com/pipermail/support-list_support.elphel.com/attachments/20111129/ac49f453/attachment-0002.html>