Local scale, rotation and position invariant word detection for optical character recognition

摘要

A system and method using a text extraction application for identifying words with multiple orientations from an image are described. The text extraction application receives an input image, generates progressively blurred images, detects blobs in the blurred images, outputs ellipses over the blobs, detects a word in the input image, orients and normalizes a first version of the word, generates an inverted version of the word, performs OCR on the first version and the inverted version of the word, generates confidence scores for the first version and the inverted version of the word and outputs text associated with the word.