Analysis and image conversion based on this analysis to improve OCR results

Question

I have an OCR project, but it only works with images in which the text is pretty straight and not upside down. (not rotated text) Therefore, I want OCR to be able to recognize any image, even inverted. But I do not know what approaches to solving this problem.

I need something like a string analysis of letters, but even then I can’t determine if the line is upside down or not.

Answer 1

If the images that you perform OCR are in a magazine or book where there is a lot of text on several lines, I suggest trying to find the page turn.

, - hough . , . "", .

hough, (, ) .

Answer 2

, , . , , . , , - .

Neocognitron -

Answer 3

, OCR . OCR , ( 90 ) . , GlyphReader OCR Engine .

, , . , , , , , .

, , , .

90 270 hough , . , , .