Handwritten Chinese Character Recognition And Aesthetic Grading Based On Deep Learning

Humans often need to use computers to process a large amount of text information in our work and life.Some of them are handwritten Chinese characters.We can greatly improve your work efficiency by using computer technology to identify handwritten Chinese characters quickly and accurately.In addition to judging the accuracy of Chinese character writing,the degree of writing aesthetics is also a very important part of enjoying handwritten Chinese characters.Learning how to write Chinese characters is a very important lesson for every Chinese child in their lives.And it is very necessary to do some research on the aesthetic evaluation of handwritten Chinese characters from the multiple meaning of the teaching needs of Chinese characters writing and the requirements of civilization inheritance.However,the current method of manually scoring Chinese characters has the defects of different evaluation standards and high work costs,which cannot meet the needs of users.The purpose of this thesis is to explore the techniques of handwritten Chinese character recognition and aesthetic grading.Combined with the latest deep learning theory,a practical handwritten Chinese character recognition and aesthetic grading system is constructed.This system can promote Chinese writing teaching and help children improve the quality of Chinese writing.The main work of the thesis is as follows:(1)We collected the handwritten Chinese character picture data set under the new natural scene—the Xiaoxuebao data set for the application scenarios and model training needs of this thesis.The data set consists of 22,050 pictures taken under natural scenes,which contain 1087 different Chinese characters.For each handwritten Chinese character,it is marked with its meaning and written aesthetic score information.(2)We proposed a handwritten Chinese character aesthetics grading method based on similarity search strategy for the demand of handwritten Chinese character aesthetic grading.Different categories of similarity search are divided according to two conditions:Chinese characters and aesthetic scores.We combine the CNN features extracted from the handwritten Chinese character deep learning network with the traditional structural features innovatively,which improved the recognition and aesthetic grading accuracy of handwritten Chinese characters effectively.(3)We built a practical handwritten Chinese character recognition and aesthetic grading system for the engineering application needs of this thesis.The model is mainly divided into three parts:handwritten Chinese character detection,handwritten Chinese character recognition,and handwritten Chinese character aesthetic grading.For these three parts,the appropriate deep learning network models were trained and optimized based on the application scenarios.The functions of the system are:firstly,the area of the Chinese character is automatically detected from an original picture and is divided into individual characters;secondly,the content of the Chinese character is automatically recognized for each Chinese character area,and finally,the system will show the writing aesthetic score of each recognized Chinese characters.We tested the system and each module,and analyzed the experimental results.The experiment verified the effectiveness of the system's recognition and aesthetic grading for handwritten Chinese characters in natural scenes.