portfolio of
Carl Dixon

Part of speech tagger

This light-weight part of speech tagger achieves 90% word accuracy on test data. It does this using a small dictionary and a bi-gram of tag transition probabilities.

Training was performed with Wall Street Journal data from the CoNLL shared tasks.

The dictionary contains 23,500 words from various word frequency lists (mostly on Wiktionary), with tags taken from the training data and entered manually.

Pronounce

A rule based g2p sketch. Achieves roughly 30% whole word accuracy and 75% syllable accuracy on CMU pronouncing dictionary data.

Bokeh

Experiments with bokeh effects.

LocationVis

Visualise a KML file exported from Google Latitude.

Metaball fluid

A cheap 2D fluid effect using metaballs and Perlin noise.

Summed Area Table

A Summed Area Table is used to apply a box filter to an image. The benefit of using a SAT is that the processing time is independent of the size of the filter kernel. Move the mouse to change the kernel size.

The algorithm is entirely stolen from this paper, although any flaws in the implementation are entirely my own:

Crow, Franklin (1984). "Summed-area tables for texture mapping". SIGGRAPH '84: Proceedings of the 11th annual conference on Computer graphics and interactive techniques: 207-212.