* Re: Any project to offer for summer code Finland
2009-01-26 17:36 Any project to offer for summer code Finland Antti Palosaari
2009-01-26 18:50 ` Anssi Kolehmainen
@ 2009-01-26 18:59 ` kilgota
1 sibling, 0 replies; 3+ messages in thread
From: kilgota @ 2009-01-26 18:59 UTC (permalink / raw)
To: Antti Palosaari; +Cc: linux-media
On Mon, 26 Jan 2009, Antti Palosaari wrote:
> Hello
> Is there any good project idea for summer code?
>
> coss.fi offers Finnish students paid summer job for open source projects.
> Project duration is three months of work.
>
> I wonder if there is some tools or libraries missing from dvb-apps example.
>
> Antti
> --
> http://palosaari.fi/
> --
Hello,
I only subscribed recently to the v4l list, myself. I have long-time
interests in digital camera support for Linux, and I have been a long-time
participant in the Gphoto project. I have written several camera driver
libraries for libgphoto2 and continue to maintain them. In my opinion,
there are several things which are badly needed. Some of them may not be
feasible topics for a summer student project, but on the other hand some
of them may. I can suggest two items.
1. The biggest problem in USB digital camera support is to be confronted
with a proprietary data compression algorithm about which no information
is easily available. Any progress with this problem would be helpful. For
example, work on the decompression algorithm for one of several cameras
which need this work done right now. However, a better topic for a student
project might be the development of some kind of toolkit which would
streamline the work on such a vexing problem. Probably, nobody will want
to tackle this one, but I remind whoever reads this that the problem is
very much present.
2. As related to V4L camera support in particular, there is the problem of
Bayer demosaicing. The current state of affairs is that raw bitmap data is
interpolated using the old method of bilinear interpolation. The results
are sometimes (quite expectedly) very ugly, with lots of artifacts at
edges in the output. Better methods exist for use in still cameras. For
example, in libgphoto2 there is a standard method which involves bilinear
interpolation with minor improvements, involving a simple edge detection
scheme. I was responsible for putting that edge detection scheme into the
code, several years ago. I was also responsible for putting in the code a
new algorithm, which is based on a simplification of the well known
Adaptive Homogeneity-Directed algorithm. This code, found in libgphoto2 in
the file ahd_bayer.c was written last year as a joint student project with
a student named Eero Salminen at Helsinki University of Technology. Eero
was studying under professor Jorma Laaksonen.
Now, the problem for using the results of last year's work for libgphoto2
in V4L is the obvious issue of speed. The code for libgphoto2 is intended
for still cameras, and the relative slowness of the code does not affect
the question of frame rate for streaming. But for a videocam or a webcam
this is an obvious problem. At the same time, it appears to me that there
ought to be a way to improve the Bayer demosaicing code in the V4L
libraries so that it gives better results without such a hit on speed. For
that matter, I am sure that the code in libgphoto2 could be improved in
this respect, too. The speed is not a factor there, to be sure, but it
might still be helpful to make it faster and, at the same time, to reduce
the memory footprint of that code. A great portion of our joint efforts
last year was in the direction of doing just that, but further progress is
still possible.
I would be very glad to work with someone on a project involving either
item 1 or item 2. The idea of such work being related to academic work and
research also appeals to me very much, as you can see below.
Theodore Kilgore
Department of Mathematics and Statistics
Auburn University, Alabama
^ permalink raw reply [flat|nested] 3+ messages in thread