Sunday, February 01, 2004

At last some good AI to get my teeth stuck in to. I've needed an OCR (Optical Carachter Recognition) component for some time, to add certain features to Read-e.

I found a royalty free bit of technology that did the job, but after 2 weeks of trying to get threw to their tech support, I've decided to make my own. Again I find myself using the concepts I first learnt about in my Physics Msci.

This is farely off the topic of dyslexia, although it may be an example of dyslexic lateral thinking.

I've gone a new route to OCR. Probability density clouds, for each carachter of each font. Similar to a weighted perceptron (a clever algorithm that atributes levels of importance, to carachteristics withinin data), a PDC is the mathematical model used to describe the "fuzz" of electrons around an atom. Its useful because it works in a dimention of probability, lending predictive modeling to the arena of "dirty data" (data containing information that is randomised). Instead of using the maths to predict the likelyhood of electron energy degredation or escape during a colission, I'm using it to predict the presense of a particular letter, in a particular font at a particular place in a image.

I hope to have a working prototype up and running within the week.

I'll keep you all informed. Mean while keep an eye on www.dyslexia-software.com

This page is powered by Blogger. Isn't yours?