I am asked to get a few dozen images containing texts in a curved form like the image below
My job is to first erase the original text from the image and then add synthetic texts in the same curved form like
Then I have to build two models
- A model to detect Urdu text's position in the image
- A model to recognize that text
I am also asked to use multiple fonts for making synthetic images like Naskh, Nastaleek, Naqsh etc
Main task is to generate around half a million synthetic images usig this format
CNN Based multi-lingual signboard text detector and recognizer
- Generate a few hundred synthetic images by hand
- Look for a way to first detect the text's position in the image frame using some object or text localization model
- Then get a model for recognition of that text
- Finally finish generating those 500k images