Character recognition method for enhancing attention mechanism by fusing multilayer features

Abstract

The invention relates to the technical field of optical character recognition in computer vision, and provides a character recognition method for enhancing an attention mechanism by fusing multilayer features. The method comprises the following steps: selecting a training picture; extracting picture features; constructing a feature fusion matrix and fusing multilayer features; carrying out feature fusion by using the associated features to enhance the feature expression ability; carrying out sequence modeling on the fused features; performing probability prediction on the features after sequence modeling; in the training stage, adopting back propagation to update the parameter weight of a network model, and obtaining a standard network model capable of being used for character recognition; in a test stage, inputting a to-be-recognized picture into the trained network model, recognizing the model, and outputting characters in the picture. According to the method, the features extracted from each level of the neural network are mutually mapped, so that the expression ability of the features is improved, and the accuracy of character recognition is improved.

Description

technical field [0001] The invention relates to the technical field of optical character recognition in computer vision, in particular to a character recognition method that integrates multi-layer features to enhance attention mechanism. Background technique [0002] In the era of the mobile Internet, a large amount of picture data can be sent and received every day, many of which contain text information, and it is particularly important to be able to accurately extract the text information in the pictures. People may need to convert the manuscripts taken by mobile phones into electronic versions, or they may need to save the text in the pictures they usually see, and so on. As the number of pictures increases, the text in the pictures also increases, and it has gradually become a new trend to be able to accurately recognize the text in the pictures. Text recognition is mainly to process the part of the picture with the text area, convert the color information in the pictu...

AI Technical Summary

Problems solved by technology

The advantage of this method is that the recognition accuracy is high; but due to its complex network structure, the recognition speed is slow

[0006] Although with the development of deep learning, the accuracy of text recognition under the action of neural network is getting higher and higher, but because the effect of recognition is closely related to the structure of the network, different network structures have different effects on the features extracted from the same picture. are quite different, and therefore give different results

Especially when the network structure is particularly complex and the number of network layers is deep, because the extracted features are too abstract, the accuracy of the final prediction result is actually lower than other methods.

Images