Optimize OCR module specially with Arabic

[samehissam]

1- Comment code using java doc. 2- What is the main problems that take much time in execution. 3- Try to solve this problems with available code. 4- If no change in performance change the available code with other that achieve your goal. 5-Try to solve the problem of the text font that you want to get from the image as it works good with text font more that 10.

[Mahmoud-zahran]

[Mahmoud-zahran]

OCR Arabic :- In the demo above, only one point-cloud template is loaded for each of the 16 gesture types. we can add additional templates as we wish, and even define our own custom gesture templates.

the algorithms used in this demo ( a straightforward algorithm called $P that represents and recognizes stroke gestures as point clouds)

• now i try to recognize letter using picture to be ensure the quality of this algorithms of detecting.
• after that => i will try to find how to get XML file representation using on detecting operation.
  1. i will try to do algorithm convert any image to XML representation .
  2. i will try to find a method to convert any font to the XML file representation.

[Mahmoud-zahran]

steps to solution.

when trying to recognize letter using picture to be ensure the quality of algorithm of detecting .

hint:- point in the algorithm consist of (x,y,stroke id) .

1. get all pixel from bitmap image using width and height .
   • [x] get the pixels using 2 loops (x of pixel width * y of pixel height) .
2. get color of each pixel to can detect the letter shape .
   • [x] get the color of pixel using getpixel(x of pixel width ,y of pixel height) .
3. detect the letter shape from background .
   • [x] convert image to monochrome (0,1) image to can
   • [x] get the point(x,y) of each white pixel or black .

[Mahmoud-zahran]

steps to solution.

second step

algorithm convert any image (the letter points) to XML representation.

1. In the $p Algorithm we have a class GestureIO has two method ReadGesture() and WriteGesture() .

2.write images of letters as a Gesture letter by letter.

• read all pixels of image of the letter .
• define the letter from background .
• define only the pixels represent the letter as a cloud of points .
• get(x,y) for each point or (pixel in the letter) .
• convert to XML using WriteGesture() method .

Notes

• use number (2) to define new letters and use it to convert the input of letter to can compare with training set and detect the leter

[Mahmoud-zahran]

the white points on the letter detected as a background and detect the letter like this .

[Mahmoud-zahran]

OCR problems.

• segmentation..................................(not finished exactly).
• Classification ................................(done).

Segmentation.

• After research for lot of 2 months for How to segment Arabic word into characters to classify char by char for High performance.
• I noticed that there is no clearly method in 8 Master research and OCR books.
• So I decided that after asking Dr.Ahmed sallam to skip this phase on OCR phases and depends on powerful of classification we can only segment the word and doing classification for word (Word by Word).
• We will show test for very similar words on classifier and its results to show classifier power.

[Mahmoud-zahran]

Solution Add all small Arabic words can be segmented as (connected word) for our XML dictionary and the classifier can distinguish between words and choose the best one.

classifier power.