On Mon, 17 Jan 2000, Jay Cox wrote:

> Kelly Lynn Martin wrote:
> Actually the first thing this plugin does is copy the pixel data into it's own
> process so there is no extra data thrashing going on (though it may be able to
> benefit from a call to gimp_tile_cache_ntiles).  Copying all the pixel info
> locally isn't exactly something that we should encourage plugins to do, but given
> the multipass nature of this plugin it seems a reasonable thing to do.
> The slow part of the plugin is finding the best matches between the input image
> and the corpus image.  I might suggest using a BSP tree or a multidimensional
> hash to speed that part of the process up, though I'm not sure if either of those
> would really be faster than the current qicksort like algorithm.  One other
> possible optimization would be to disregard duplicated patches from the corpus
> image.

Yep, the corpus search is what is slowing it down, my current algorithm is
pretty ugly -- a combination of a binary search tree and k-means.
Hopefully i can improve it for the next version.

btw, thanks to all the people who replied to my question about patents.

Paul Harrison

Web:   http://yoyo.cc.monash.edu.au/~pfh/
                               grey spiderwebs define space

Reply via email to