This project creates synthetic hebrew characters that are broken by overlapping Arabic characters. The aim is to train a deep learning network for restoring broken characters into their original forms.

Creates root Hebrew letters of 8 different fonts and root Arabic letters of 8 different fonts. Using these root letters creates homographic versions of root Hebrew letters and root Arabic letters in the number of a limit set by the user. For every Hebrew letter an Arabic letter is randomly selected and randomly overlapped on this Hebrew letter. Overlapping letter parts are deleted to handle a broken Hebrew letter.

Following images show some original Hebrew letters, Hebrew letters overlapped by Arabic letters and the final broken Hebrew letters.

Naming convention for the letter images is: labelnumber-samplenumber.png

Label number of Hebrew letters are mapped in hebrew-letters.txt file.

Label number of Arabic letters are mapped in arabic-letters.txt file.