(no subject)

[njordsifusansoo]

I hope this function I just finished coding has no errors...I am building a binary tree for a complicated data structure level by level (instead of say recursively). I am doing it level by level because I need the option to limit the number of leaves. The data structure is complicated because I am sorting high dimensional points (12+ dimensions) by which point they are closer to of the centroid and a perturbed centroid (a point just a little away from the centroid). Its even more complicated because the high dimensional points are actually the neighborhoods of pixels (well a causal nieghborhood and if I have a gaussian pyramid its corresponding noncausal neighborhood with each pixel from both being a dimension).

The binary tree is a Tree Structure Vector Quantization (TSVQ) for accelerating the neighborhood search for Fast Texture Synthesis as detailed in this paper(pdf link). The point of it is so you just have to search the tree generated off of your sample texture instead of doing an exhaustive search over your entire sample texture (if your sample texture is say 64x64, that is 4096 comparisons in an exhaustive search per pixel in the output texture). If it works, it should reduce computation time from around an hour (for a okay sized output texture using a gaussian pyramid) to under a minute with minimal lost in quality.

Comments

[darkamdusias.livejournal.com]

Wow, that sounds awful. There's probably a cream for that out there somewhere, though. I doubt that you're the first person who's ever had a swollen perturbed centroid.

;-P