[MPlayer-users] Improved New Edge-Directed Interpolation

Nikolaj Poroshin porosh3 at psu.ru
Sat Aug 11 21:22:50 CEST 2007


Guillaume POIRIER <poirierg <at> gmail.com> writes:

> 
> Hi,
> 
> On 27 Dec 2006 15:30:42 -0000, p8kdnvz02 <at> sneakemail.com
> <p8kdnvz02 <at> sneakemail.com> wrote:
> > Hi,
> >
> > I just found this algorithm, that produces very nicely scaled pictures. 
Screenshot aswell.
> >
> > www.mathworks.com/matlabcentral/fileexchange/loadFile.do?
objectId=13470&objectType=FILE
> >
> > Could it be something for mplayer?
> 
> All it takes is having someone re-implement it in C and add it to swscaler :)
> 
> Guillaume

I've made implementation of it + some minor improvements to check how it 
performs on variety of images. It seems that method is performs better than 
typical lancroz3 upscale only on contrast diagonal/sloping edges, where lancroz 
(and others) produces stairs-effect artifacts. On any other regions (most part 
of an image) i have to fall-back to lancroz3 upscale to not lose image 
sharpness. Lancroz3 upscale performs better than this method on typical 
"texture" parts of an image.

Here's the links to the picture i've found with most noticeable differences 
http://s1.imgdb.ru/2007-08/11/pent0-PNG_qfx57e6z.png
http://s1.imgdb.ru/2007-08/11/pent-lanc-PNG_sano8a72.png
http://s1.imgdb.ru/2007-08/11/pent-inedi-PNG_da3g3qab.png
(also original and lancroz resized one)

Here's another picture (take a note on those looong straps ;-) )
http://s1.imgdb.ru/2007-08/11/j0-PNG_bdb3ro2r.png
http://s1.imgdb.ru/2007-08/11/j-lanc-PNG_5sf8q79d.png
http://s1.imgdb.ru/2007-08/11/j-inedi-PNG_xxq7kano.png

So, I'm thinking - could it be something for mplayer/etc or don't bother about 
it?
Pls say that you think? Maybe you have another good real picture to test, on 
which lancroz3 upscale produces noticeable artifacts?







More information about the MPlayer-users mailing list