neural network - How do I make an OCR Program? -
i want make program takes image input , outputs text. know can use neural network turn image of single character character. difficult part is: given image text in it, how produce rectangles around each individual character?
so sort of methods used , know of research papers discuss how it? thank you
a basic approach make histogram of black pixels. first: project pixels on line. deep valleys in histgram indicate separation between lines (try different angles if paper might tilted). then, per line (or per page if know font monospaced) project pixels on horizontal histogram. give strong indication of inter character spaces. minimum gives value average character height , width in next steps.
after that, need take care of kerning (where characters overlap). find connected pixels, possibly first doing dilatation or erosion on image compensate scanning artifacts.
depending on quality of scan image may have use more advanced techniques, going.
Comments
Post a Comment