Hi(gh)!

On 19.04.2013 00:04, Urs Holzer wrote:

> First of all, doing something like this is a huge amount of work never
> to be underestimated.

Yes, we speak of currently about 18,500 image files...

> I find the idea intriguing and I like to share a
> few ideas of my own regarding this.
>
> Let me first extract some keywords from your post:
> * publicly accessible database
> * lots of metadata for every image/post
>    - extractable from the images:
>      size, ratio, color distribution

I'm working on a small C program running on the console which will be 
able to do this - at least after having the image file converted to 
uncompressed TGA!

>    - extractable from scene files:
>      used features, POV-Ray version

...or, if no scripts are available, probably from the POVer's posting on 
p.b.i!

>    - not easily extractable:
>      topics, keywords for objects shown in the image
> * Connecting images to scene files (linking p.b.i posts to p.b.s-f
>    posts)
>
> I would recommend the following strategy: Intertwine the metadata with
> the semantic web. This gives us a plethora of keywords. For example,
> look at DBPedia. It provides identifiers for everything described by an
> article on Wikipedia. Yes, POV-Ray too:
> http://dbpedia.org/resource/POV-Ray

I looked it up... appears both vague and complicated to me! And as you 
say it was still in an experimental stage, I think a relational keyword 
table (or better, several tables making up a whole keyword hierarchy) 
would be far more reliable.

> Clear drawback: The semantic web is kind of new territory while the
> usual relational database with a web interface is robust and proven.

I think so!

See you in Khyberspace!

Yadgar

Post a reply to this message