| View previous topic :: View next topic |
| Author |
Message |
dondora
Guest
|
 Posted: Tue Oct 07, 2008 11:18 am Post subject: Re: searching faces using pca |
 |
|
Here it is.
http://mercury.jnu.ac.kr/~koninja/pca.zip
I>m now scrutinizing what you wrote.
Thank you so much. |
|
| |
|
 Back to top |
illywhacker
Guest
|
| Posted: Wed Oct 08, 2008 5:46 pm Post subject: Re: Has any made the switch from matlab to scilab (or other |
|
|
On Sep 26, 1:17 am, ImageAnalyst <imageanal...@mailinator.com> wrote:
[quote]On Sep 25, 2:55 pm, Bryan Heit <bryans.spam.t...@gmail.com> wrote:> Thanx everyone for you reply. I>ll repost this on the matlab group,
but it sounds like this may be a bust...
Bryan
----------------------------------------
Bryan:
Don>t expect much sympathy or help there. This topic generally comes
up about once a year there -- people name a few packages and lots of
people bash them and point out their deficiencies, even though
everyone there knows that MATLAB has it>s quirks and limitations (such
as no user interface for image processing functions, just command line
interface, and a lousy dialog box editor program).
[/quote]
Yeah, but bad as that is, the alternatives, although free, are more or
less unusable, sad to say.
illywhacker; |
|
| |
|
| Back to top |
illywhacker
Guest
|
| Posted: Wed Oct 08, 2008 5:59 pm Post subject: Re: Mexican hat |
|
|
On Sep 26, 1:23 am, ImageAnalyst <imageanal...@mailinator.com> wrote:
[quote]On Sep 25, 10:44 am, illywhacker <illywac...@gmail.com> wrote:
On Sep 25, 3:36 pm, Anicet Franken <Anicet.Fran...@tata.titi.toto
wrote:
Hello,
who knows the following kernel : mexican hat?
Normally this is the negative of the Lapacian of a Gaussian. As
always, try wikipedia yourself before you bother people here:
http://en.wikipedia.org/wiki/Mexican_hat_wavelet
illywhacker;
illywhacker:
I didn>t know that - learn something new every day. I don>t use
wavelets but have a background in optics and imaging and there the
Mexican Hat "normally" refers to the Bessel function, like Jomar
said. Just as often, or maybe more or less, it>s also called the
"Sombrero function." You get this from the diffraction pattern of a
circular aperture - analogous to the sinc function which you get from
diffraction of a rectangular aperture. You can convolve your image
with these to simulate blurring by imaging through apertures shaped
like that.
[/quote]
Yes, thanks: I see that. I guess the truth is that it could be used to
refer to any function of radius with a positive centre and negative
surround, tending to zero at infinity. The Bessel function has
multiple zeros, which is unlike any sombrero I have ever seen, but
then sombreros do not extend to infinity either. Calling it a kernel
or not is irrelevant though. The Laplacian of a Gaussian is also a
kernel.
illywhacker; |
|
| |
|
| Back to top |
bolme
Guest
|
| Posted: Thu Oct 09, 2008 5:35 pm Post subject: Re: Has any made the switch from matlab to scilab (or other |
|
|
Bryan,
I am also a little late. I have been very happy with python/scipy. I
switched to python about 5 years ago and I have been very happy with
it. I have found that scipy can handle about 95% of the features that
I appreciated from matlab and many of the functions even have the same
name and similar interface. Python was also pretty easy to learn. A
year ago I tried matlab again and decided that I did not want to go
back.
Another benefit of python is that the language is well developed and
is much more general purpose than matlab. For example you can build
large object oriented systems with python that would not be possible
with matlab, and you have access to a huge library of tools that you
can use to easily solve many problems. I have also found that python
has interfaces to many computer vision and machine learning libraries
that I use in my research.
The disadvantages of python is that it is not a "research language"
and it does not have the same commercial support as matlab. This
means that there is not a clean nice gui, the plotting functions are
not as good, and it does not have as many specialized toolboxes and
functions as matlab.
I gave a talk summarizing computer vision toolkits a few weeks ago and
it included python and matlab. Slides can be found here:
https://www.cs.colostate.edu/wiki/mediawiki/images/e/ef/CVToolsv2.pdf
Hope this helps,
Dave
---
http://www.cs.colostate.edu/~bolme |
|
| |
|
| Back to top |
Milind Joshi
Guest
|
| Posted: Thu Oct 09, 2008 8:25 pm Post subject: Re: Has any made the switch from matlab to scilab (or other |
|
|
On Oct 9, 1:35 pm, bolme <bo...@cs.colostate.edu> wrote:
[quote]Bryan,
I am also a little late. I have been very happy with python/scipy. I
switched to python about 5 years ago and I have been very happy with
it. I have found that scipy can handle about 95% of the features that
I appreciated from matlab and many of the functions even have the same
name and similar interface. Python was also pretty easy to learn. A
year ago I tried matlab again and decided that I did not want to go
back.
Another benefit of python is that the language is well developed and
is much more general purpose than matlab. For example you can build
large object oriented systems with python that would not be possible
with matlab, and you have access to a huge library of tools that you
can use to easily solve many problems. I have also found that python
has interfaces to many computer vision and machine learning libraries
that I use in my research.
The disadvantages of python is that it is not a "research language"
and it does not have the same commercial support as matlab. This
means that there is not a clean nice gui, the plotting functions are
not as good, and it does not have as many specialized toolboxes and
functions as matlab.
I gave a talk summarizing computer vision toolkits a few weeks ago and
it included python and matlab. Slides can be found here:https://www.cs..colostate.edu/wiki/mediawiki/images/e/ef/CVToolsv2.pdf
Hope this helps,
Dave
---http://www.cs.colostate.edu/~bolme
[/quote]
I generally agree with many of the points-of-view here, and there are
good and bad things about both commercial or open source.
My biggest problem with MATLAB is that you cannot get to any kind of
real working software once you>ve finished the prototyping... so, if I
want to give a tool to a colleague or associate or customer or partner
or anyone who does not have a MATLAB license, they cannot do anything
with it unless they have a working copy of MATLAB.
I>ve found that it pays to look out for tools that you need
specifically. Yes, MATLAB has some great built-in toolkits and you can
get a lot of code from other users, but that is true of many,many open
source packages.
Eventually, it boils down to one>s own comfort level...I can imagine
that Mathworks has, by a big margin, more developers, testers,
documentation makers, evangelists, and support personnel than any one
open source project.
So it is a matter of time, but it is a chicken-and-egg, don>t be
afraid to look out there for other alternatives to MATLAB... for most
researchers, MATLAB was what they were told to use in school, it was
the tool they learned, and probably the only environment they are
comfortable in, so if you ask them to switch, many will complain very
loudly.
If you want to change that, suggest using an alternative tool in the
undergraduate courses... :-) Most universities get into the special
low-priced student licensing offered by Mathworks... maybe not now,
but in a few years, there could be some real choices, even for those
addicted to MATLAB!
Best Regards,
Milind Joshi
IDEA TECHNOSOFT INC.
http://www.ideatechnosoft.com |
|
| |
|
| Back to top |
Chris Maryan
Guest
|
| Posted: Fri Oct 24, 2008 4:22 pm Post subject: Re: Splitting concave figures. |
|
|
On Oct 24, 6:57 am, A n g l e r <p|k|o|n|i|u|...@h-o-t-m-a-i-l.c_o_m>
wrote:
[quote]Dear all.
Imagine yourself a binary snap of a boomerang or a propeller or any
other figure which is partially concave. I wonder how I could split it
onto convex objects. OK, they don>t have to be convex, though the
original image should be split at the most concave coordinates that
affect the overall shape the most. There definitely will be some noise
as well, therefore a small concave bit shouldn>t give a rise to
splitting, only big salient bits. Any ideas? Any publications or
well-studied algorithms you>re aware of? Speed is kind-of crucial.
Cheers in advance,
Peter.
[/quote]
Knowing nothing about the topic, but eager to take a guess, how about:
1. Calculate the convex hull of the object
2. from the convex hull outline, measure the perpendicular distance to
the object outline (perpendicular wrt the convex hull)
3. Where the distance is greatest, that>s probably a good point to cut
the object. (i.e. cut along to any other identified maxima, or to the
opposite side of the object.
4. Repeat for any of the generated sub objects until they are all
convex.
Chris |
|
| |
|
| Back to top |
Frank Iannarilli
Guest
|
| Posted: Mon Nov 03, 2008 2:19 pm Post subject: Re: Splitting concave figures. |
|
|
http://codesuppository.blogspot.com/2006/04/approximate-convex-decomposition.html
Treats 3D polyhedra, but you can game it to work for your 2D problem,
I suppose.
Cheers,
Frank |
|
| |
|
| Back to top |
|
