CODEWITHSUNDEEP
neural-networkocrpattern-recognition
Romantic Electron
Romantic Electron

Reputation: 779

OCR for different sizes of text with NN?

How to deal with recognition of a pattern where the input character font size in the image is different in size than the those it was trained with? How do I input this to my trained neural network?At the moment I just know about feed-forward neural networks and I assume they are sufficient for it.Do I need something else?

Upvotes: 1

Views: 889

Answers (1)

Denis Tarasov
Denis Tarasov

Reputation: 1051

Most practical systems do some preprocessing like edge detection followed by characted extraction and character scaling (see this paper for example). Thus, all characters are scaled to the same size. One can try other approaches, for example training NN to recognize some small window as part of character or not part of character (while given larger context window), and then merging adjacent parts (like its done with face recognition). For more info on this topic see this paper that discusses possibilities for text detection with NN without preprocessing

Upvotes: 3

Related Questions