Text recognition method and device

Abstract

The embodiment of the invention relates to the technical field of information recognition, and provides a text recognition method and device, and the method comprises the steps: firstly obtaining a to-be-recognized text image; and then inputting the to-be-recognized text image into a text recognition model to obtain text information in the to-be-recognized text image output by the text recognitionmodel. Wherein the text recognition model is constructed based on a convolutional neural network, the convolutional neural network comprises a splicing layer, and the splicing layer is used for splicing at least two items of input data of the splicing layer; wherein the input data and the output data are both fixed-point data. The output data of the splicing layer correspond to a predetermined first type of fixed-point coefficient, each piece of input data of the splicing layer corresponds to a predetermined second type of fixed-point coefficient, and the first type of fixed-point coefficientis equal to the second type of fixed-point coefficient. At least two items of input data of the splicing layer can be directly spliced, and the splicing operation is simplified.

Description

technical field [0001] The present application relates to the technical field of information recognition, in particular to a text recognition method and device. Background technique [0002] In recent years, text recognition has attracted the interest of a large number of researchers. Thanks to the research of deep learning and sequence problems, many text recognition techniques have achieved remarkable success. [0003] Currently, text recognition is mainly aimed at recognizing text in images. When recognizing text in an image, it is usually implemented using a neural network model. However, in the process of text recognition using the neural network model, there are a large number of complex data calculations, such as the calculation between the floating-point model parameters of each network layer and the image data that may be in decimal form, resulting in the occupation of text recognition. A lot of computer resources and a long time consumption. Moreover, when using...

AI Technical Summary

Problems solved by technology

However, in the process of text recognition using the neural network model, there are a large number of complex data calculations, such as the calculation between the floating-point model parameters of each network layer and the image data that may be in decimal form, resulting in the occupation of text recognition. A lot of computer resources, long time consumption

Moreover, when using the neural network model to process tasks, it is often necessary to rebuild the neural network model. At this time, the complexity of parameter processing for each network layer will lead to low efficiency in building the model, which in turn greatly reduces the cost of the neural network model for text recognition. efficiency

[0004] To sum up, when using a neural network model for text recognition, the process of constructing a neural network model using parameters of complex data types is complex and takes up a lot of resources, and the constructed neural network model takes up a lot of computing resources and consumes time when processing tasks. Long, inefficient processing tasks

Images