OPTICAL CHARACTER RECOGNITION BY MEANS OF COMBINATION OF NEURAL NETWORK MODELS Russian patent published in 2022 - IPC G06V10/40 G06V30/18 G06N3/02 

FIELD: physics.

SUBSTANCE: invention relates to a text recognition method and system. In the method, a computer system obtains an image with text; selecting, using the feature extraction unit, a plurality of features from the image; applying a first decoder to a plurality of features to generate a first intermediate output, wherein the intermediate output is a hypothesis of a sequence of text characters; applying a second decoder to a plurality of features to generate a second intermediate output, where the feature extraction unit is shared by the first and second decoders; determining, based on the language model, a first quality metric value for a first intermediate output and a second quality metric value for a second intermediate output; and in response to determining that the value of the first quality metric is greater than the value of the second quality metric, selecting a first intermediate output for presenting the text.

EFFECT: high efficiency and accuracy of text recognition.

20 cl, 9 dwg