Text recognition model training method and device and text recognition method and device

Abstract

The invention provides a text recognition model training method and device and a text recognition method and device.The text recognition model training method comprises the steps of inputting an obtained first sample image into a region detection network of a text recognition model, and determining at least one first sub-sample image of the first sample image; inputting the at least one first sub-sample image into a character recognition network of a text recognition model, and determining a prediction text of the first sample image, the prediction text being determined based on a sequence relationship between image features in each first sub-sample image extracted by the character recognition network; determining a first error value between the predicted text and a first sample label of the first sample image, wherein the first sample label is a text actually existing in the first sample image; and training the text recognition model based on the first error value until a training stop condition is reached. According to the method, the accuracy of the obtained predicted text is higher, and the training efficiency of the text recognition model can be improved.

Description

technical field [0001] The present application relates to the technical field of image processing, and in particular to a text recognition model training method and device, a text recognition method and device, computing equipment, and a computer-readable storage medium. Background technique [0002] Images can sometimes contain rich text information, and text recognition for images is to obtain the text in the image by performing certain image processing on the image. Usually, image text recognition can include two stages of area detection and character recognition. In the area detection stage, the area where text exists in the image is detected. In the character recognition stage, the area where text exists in the image is recognized to obtain the text in the image. text. [0003] In the prior art, the character recognition stage of text recognition is usually implemented through a character recognition network. Moreover, the existing character recognition network usuall...

AI Technical Summary

Problems solved by technology

However, CTC loss only focuses on the characteristics of a single character, and the accuracy of the recognized text is related to a single character and the semantic relationship between characters, so the text determined by the above method may be inaccurate or semantically incoherent. Accurate text training a text recognition model may reduce the training efficiency of a text recognition model

Images