# Text image analysis with the haar wavelet

Discussion in 'Java' started by tommygun101, Jun 8, 2007.

1. ### tommygun101Guest

well i havent made any algorithms as of yet, but i know how to go a
bout it.
the only wavelet code i have done is in mathematica, not java.

You would basically convert a the bitmap image of the text into a nxm
matrix.
Then you would analyse the vertical and horisontal "lines" as a
discrete function,
with each color coming through as a frequency.

So the spaces between each letter would have only 1 frequency
"corresponding with white or whatever"
and you could then single out each letter and analyse it against a
database of known frequencies.
Then using some stochastic process you could estimate the closest
letter it coresponds to.
(aside: the FB! actually uses the haar wavelet to store fingerprints
in a compressed form)

You could also analyse each line as a combination of frequencies, but
then you would need to
make a database of every possible charater combination, which is not
feasable. it would increase
at a rate of n!... which is in NP range.

There are many other easier, ways of analysing computer text in a
image, im sure, just coming from the
applied math field, im not too sure of any other methods of analysing
text. If anyone could give me alternative
methods i would apreciate it

I think the biggest advantage of haar wavelet is it can differentiate
other things in an image if you as the programmer know the structure
of the shape you are looking to isolage.

P.s . I know this is a bit off topic of java programming but its a
possible algotithm aproach, dont get too angry

What ideas do you have?

tommygun101, Jun 8, 2007