A handwritten sample recognition method and system based on sample enhancement

Abstract

The invention discloses a handwritten sample recognition method and system based on sample enhancement, wherein the method includes: S1, generating a marked sample, marking the handwritten characters in the image sample, cutting out the handwritten characters from the image sample and classifying them; S2, sample enhancement, randomly transform the labeled samples to generate transformed samples, use the generation model to generate enhanced samples with the same distribution as the transformed samples; S3, sample synthesis, use enhanced samples to generate training samples; S4, model training, use training samples Training the detection and classification model and the handwritten sample recognition model; S5, the recognition application, using the trained detection and classification model to detect the position of the handwritten character, and then recognizing the handwritten character through the handwritten sample recognition model. The invention effectively overcomes the problems of low offline handwriting recognition accuracy, difficulty in labeling handwriting recognition samples, and slow model improvement in the prior art by increasing the diversity of training samples and optimizing the detection classification model and recognition model.

Description

technical field [0001] The invention relates to the technical field of text image intelligent recognition, in particular to a method and system for handwritten sample recognition based on sample enhancement. Background technique [0002] Text images mainly refer to documents that convert paper documents into image formats in some way. Text images may include: tables, pictures, machine-generated fonts, and even handwritten handwriting. Handwritten handwriting includes handwritten fonts and handwritten graphics. Generally speaking, optical character recognition and deep neural network recognition are usually used for image recognition of non-handwritten handwriting. For handwriting recognition, online recognition and offline recognition are usually used. The online recognition mainly refers to obtaining information such as stroke order and pressure sensitivity of handwritten handwriting, judging the handwriting content through a model, and feeding back all possible results to...

AI Technical Summary

Problems solved by technology

However, in the existing offline recognition, due to the lack of uniform standards for handwritten handwriting compared with general fonts, the handwriting styles of writers are quite different, and handwriting is scribbled and continuous strokes are common, resulting in low recognition rate and difficulty in labeling handwriting recognition samples. slow up

Images