[Gimp-developer] A GSoC Purpose: Improve the Performance of SIOX

[Sven Neumann]

Hi,

On Sat, 2009-03-14 at 11:52 +0800, Jenny wrote:

> My name is Ding Jie, and Jenny is my English name. I'm a first-grade
> post-graduate student, major in Image Processing, and very interested
> in algorithms of this field.
> 
> Thanks for neo's suggestion, I found SIOX is very interesting. I have
> read some paper about SIOX written by Gerald, and feel SIOX have
> potential to improve.
> 
> Firstly, SIOX performs not good when dealing with gray images.
> Secondly, it isn't good at extracting things from a background whose
> color is similar to foreground. I'm wondering is it possible to use
> some new approaches side-by-side, and make the final output as the
> combination of all those approaches' output.

SIOX can by definition only extract things from a background if the
foreground color is sufficiently different. After all that is how the
algorithm decides what to separate. So what you suggest here sounds more
like adding a different separation algorithm and not like improving
SIOX. But of course I might just have misunderstood you.

The SIOX tool has several areas where it is lacking. The most notable is
that it does a binary separation. Either a pixel is foreground or it is
background. This causes jaggy edges and makes it necessary to do manual
post-processing on the selection. The SIOX team around Gerald has
suggested several improvements that would help to improve this. Their
papers describe a refinement brush that should be used on the selection
boundary to improve the results. It would be nice if this could be added
to the GIMP SIOX tool.


Sven