Character detection model pre-training method and device

Abstract

The embodiment of the invention provides a character detection model pre-training method and device, and the method comprises the steps: inputting a character sample into a text encoder to obtain a character feature, inputting an image sample into an image encoder to obtain an image feature, and enabling the character sample to be extracted from the image sample; according to a data dictionary and the image features, determining whether the image sample contains a text sample, and obtaining a text containing result, the data dictionary comprising the text sample; determining a corresponding relation between the character sample and the image sample according to the character features and the image features, and obtaining an image-text corresponding result; predicting the masked character sample according to the character features and the image features to obtain a character prediction result; and according to the containing result, the image-text corresponding result and the text prediction result, performing parameter adjustment on the image encoder to obtain a pre-trained text detection model. Visual representation has semantic knowledge, so that the problems of ambiguity in rows and columns and the like caused by insufficient semantic knowledge are avoided.

Description

technical field [0001] The embodiments of this specification relate to the technical field of model training, and in particular to a method for pre-training a text detection model. Background technique [0002] With the rapid development of personal consumer electronics (digital cameras, mobile phones, etc.) , card pictures, street view videos, etc.) to extract text information in Optical Character Recognition (OCR) technology has been widely used. With the advent of the era of deep learning, OCR technology has gradually moved from simple character recognition of scanned documents to an era of solving complex text scenarios such as complex text distribution, artistic words, table text, and even handwritten formulas in a wide range of scenarios. Pan-OCR technology is usually divided into three stages: text detection, text recognition, and format understanding. In the text detection stage, it is necessary to train the text detection model based on deep learning and training ...

AI Technical Summary

Problems solved by technology

However, extracting characters from complex text formats often involves understanding the semantics of the text content (especially in Chinese scenes involving row and column ambiguity, wide-spaced words, etc.), and the distance between characters cannot simply be used to judge the line of text. basis, which in turn affects the effect of text detection

Images