I'm a very very fresh newbie in this field of programming and wondering if anyone could help me with this. I'm now doing a very begining image pre-processing where for the time being the requirement of my work need me to look for an algorithm that can segment a word to alphabet... for example:- the word "EXAMPLE" to "E","X","A","M","P","L","E" where each and every alphabet will be recognize later .. i know it is something related to word segmentation.. if the word is not distorted then it might be easy.. but how bout when the word is rotated by 20 degree... then it will be really confusing... to deskew back the image using the line detection algorithm takes lots of times ... i just hope there will be somebody could help me with this...

I do not believe I understand your question properly...

Do you want an algorithm for OCR (Optical Character Recognition)? It looks like it, but that's no beginners job.

Do you want to split up strings into characters? That should be easy enough, even for beginners.

Could you be more specific?

tia,
CipherCraft

start by knowing your enemy http://en.wikipedia.org/wiki/Captcha
and the theory behind the attacks http://en.wikipedia.org/wiki/Neural_network

Quote:Original post by CipherCraft
I do not believe I understand your question properly...

Do you want an algorithm for OCR (Optical Character Recognition)? It looks like it, but that's no beginners job.

Do you want to split up strings into characters? That should be easy enough, even for beginners.

Could you be more specific?

tia,
CipherCraft

Thanks for helping in the first hand. Actually, i just want to split the strings into character (eg. a word to it's respective alphabet or a string of number to its respective individual number), and yea it is easy enough if the string is not distorted. But when the string is rotated, then there comes in the headache... i search for few algorithm here including the line detection method using hough transfor to deskew the string that being rotated, but the computation time prove to be too expensive.. around 7 to 9 seconds. Thats why im looking for a much better yet simpler algorithm to perfrom this task. Thanks again for responding :)

Quote:Original post by spx2
start by knowing your enemy http://en.wikipedia.org/wiki/Captcha
and the theory behind the attacks http://en.wikipedia.org/wiki/Neural_network

Thanks for the heads up...

you could take in consideration also some topological properties of letters.
http://www.percepp.demon.co.uk/stpte1.jpg
(we are reffering ofcourse to the roman alphabet)
for example, the letter A is different from all the others from a topological
point of view in the sense that , whatever the rotation , it is still unique
from any other letter.
you could build up an algorithm, on recognising
continous curves(the edges of a graph),so that the letter A for
example becomes a graph with 5 nodes and 5 edges.
the structure of the graph is like this

  . /  \  .----.|    ||    |.    .

now the dots are the nodes and the lines are the edges.


so basically like that you can construct also a pattern for
the letter B wich would have ofcourse 4 nodes and 5 edges.

like this

. | \ .__. | / ./ 

and altough this later one seems a little bit different from B,
it has the same structure...and go on and describe all letters like
that,the use filling algorithms to get the edges of the image you need.
but this would probably be the hardest of all.

ofcourse these are only ideeas,but if you don't want to loose
time trying to solve the problem on your own you can see what
others have thought of with practical algorithms :
http://www.scirus.com/srsapp/search?q=pattern+recognition+algorithm&ds=jnl&ds=nom&ds=web&g=s&t=all
http://citeseer.ist.psu.edu/cis?q=pattern+recognition+algorithm&cs=1

Quote:Original post by spx2
ofcourse these are only ideeas,but if you don't want to loose
time trying to solve the problem on your own you can see what
others have thought of with practical algorithms :
http://www.scirus.com/srsapp/search?q=pattern+recognition+algorithm&ds=jnl&ds=nom&ds=web&g=s&t=all
http://citeseer.ist.psu.edu/cis?q=pattern+recognition+algorithm&cs=1

hmm.. thanks for d info.. but what actually im doing now is not pattern recognition.. instead, im doing some image matching operation:-
The criteria for this operation is tighter compare to pattern recognition... for example for the same letter of X, it can be rotated or scaled... but it will not tolerate the different type of shape although it is the same X... i have already do this matching operation by using ring projection theorem and it is practical and also fast.. but if i want to match a string of word... for example "HELLO" ... it is not rather convincing... thats why im looking for some algorithm that can separate the word "HELLO" to the respective alphabet "H"..."E"...."L"...."L"..."O".... and then only do the matching operation for each and every character.... if the picture is not rotated, then i can do d word segmentation by counting the white pixel separation between d alphabet ... if the word if rotated... then it is hard to separate it.... im thinking of using an algorithm that can find the angle of distortion to normalize it back to the orginal x-axis ... but so far the algorithm i found prove to be expensive in term of computation time... for eg... Mellin transform that invovle Fourier transform....better known as Mellin-Fourier descriptor....and also Hough Transform that detect the straight line in the alphabet before calculate the angle of distortion... Just wondering will there exist any better algorithm... Anyway ... really appreciate your help... thanks =)