Haar algorithm into CImg...

[Gilles Caulier-4]

Hi all,

i have take a look into CImg web site and seen than 1.2.7 release is out.

From the CHANGELOG file, we can read :

*-----------------------------
* Changes from 1.2.6 to 1.2.7
*-----------------------------
*-----------------------------

* New features :
     - Added functions 'CImg<T>::get_haar()', 'CImg<T>::haar()' allowing to compute the direct and
     inverse Haar multiscale transform on 2D or 3D images.

Like you know, we will store and use Haar matrix into database to perform image fuzzy searches and duplicate items).

What do you think to use this way to compute Haar matrix ?

Alternative way is to import Haar algorithm from ImgSeek :

http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgSeekLib/haar.cpp?revision=56&view=markup
http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgSeekLib/haar.h?revision=56&view=markup

... But this can be more complicated to do...

Your viewpoints ?

Gilles Caulier

_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel

[Marcel Wiesweg]

> * New features :
>      - Added functions 'CImg<T>::get_haar()', 'CImg<T>::haar()' allowing to
> compute the direct and
>      inverse Haar multiscale transform on 2D or 3D images.
>
> Like you know, we will store and use Haar matrix into database to perform
> image fuzzy searches and duplicate items).
>
> What do you think to use this way to compute Haar matrix ?
>
> Alternative way is to import Haar algorithm from ImgSeek :
>
> http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgS
>eekLib/haar.cpp?revision=56&view=markup
> http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgS
>eekLib/haar.h?revision=56&view=markup
>
> ... But this can be more complicated to do...
>
> Your viewpoints ?

If there are no strong reasons not to use CImg in this case (very slow), which
I dont believe - in ImgSeek it was written in Python with native code parts
after all - then it comes just right in time to save us some time, doesnt it?

>
> Gilles Caulier


_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel

[Gilles Caulier-4]

2008/1/28, Marcel Wiesweg <[hidden email]>:

> * New features :
>      - Added functions 'CImg<T>::get_haar()', 'CImg<T>::haar()' allowing to
> compute the direct and
>      inverse Haar multiscale transform on 2D or 3D images.
>
> Like you know, we will store and use Haar matrix into database to perform
> image fuzzy searches and duplicate items).
>
> What do you think to use this way to compute Haar matrix ?
>
> Alternative way is to import Haar algorithm from ImgSeek :
>
> http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgS
>eekLib/haar.cpp?revision=56&view=markup
> http://imgseek.svn.sourceforge.net/viewvc/imgseek/trunk/imgseek-python/imgS
>eekLib/haar.h?revision=56&view=markup
>
> ... But this can be more complicated to do...
>
> Your viewpoints ?

If there are no strong reasons not to use CImg in this case (very slow), which
I dont believe - in ImgSeek it was written in Python with native code parts
after all - then it comes just right in time to save us some time, doesnt it?

From ImgSeek, there is only haar algorithm witch have been written in C++ like you can see.

From CImg, i don't know if algorithm speed-up. And, you have right, speed must be the most important factor here...

I will make a small test program to compare...

Gilles

_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel