Research And Application Of Intelligent Calligraphy Aesthetic Evaluation Method

Calligraphy is the treasure of the Chinese nation,and its inheritance is inseparable from the popularization of calligraphy education.The basis of calligraphy learning is that students practice calligraphy copying.Teachers evaluate and guide their calligraphic copies to help them compare the differences between copies and templates to improve handwriting.However,calligraphy teachers have limited resources and are difficult to provide objective aesthetic evaluation and guidance to students,so the research of intelligent calligraphy aesthetic evaluation system is of great significance.At present,the field of intelligent aesthetic evaluation mainly focuses on photos such as landscape and people,and there is a lack of aesthetic evaluation research on Chinese calligraphy.This paper carries out an in-depth study on this subject,and the main contents are as follows:(1)In view of the current lack of high-quality aesthetic evaluation datasets on Chinese calligraphy,which is the basis of calligraphy aesthetic evaluation method,this paper establishes the Evaluated Chinese Calligraphy Copies(E3C)dataset with public aesthetic evaluation opinions based on the students’ calligraphic copies in class.The high-quality part covers8995 calligraphy images including 40 different Chinese characters.Each case has an aesthetic evaluation label in the range of 1-10 with three significant digits after evaluating by 10 experts and amateurs.In this paper,the dataset is reasonably divided and the evaluation metric is selected,and the benchmark experiment is carried out.(2)In order to merge the aesthetic rules and aesthetic prior knowledge of calligraphy,the feature engineering on calligraphy aesthetic and its aesthetic evaluation algorithm based on image processing are proposed.This paper designs 12 hand-crafted calligraphy aesthetic features from the perspective of calligraphy font,structure and stroke,and verifies the effectiveness of the features.Through the calligraphy aesthetics feature engineering,this paper proposes a calligraphy aesthetics evaluation algorithm based on machine learning with 0.989 MAE(mean absolute error)and 0.548 PCC(Pearson correlation coefficient)on the E3 C benchmark testset.Compared with the benchmark experiment with the same regressor,the MAE and PCC are improved by 0.072 and 0.111 respectively.(3)Due to the limitation of the hand-crafted calligraphy aesthetic features,hidden aesthetic information of calligraphy images is omitted,which affect the performance of calligraphy aesthetic evaluation.For this problem,based on the Siamese network architecture and transfer learning,this paper proposes the Siamese regression network(SRN)to extract the deep aesthetic representation to realize calligraphy evaluation.To increase the interpretability of the evaluation method and introduce the prior knowledge of calligraphy aesthetics,based on the hand-crafted calligraphy aesthetic features and SRN,this paper proposes the Siamese regression aesthetics fusion evaluation(SRAFE)method,and verifies the importance of its components through ablation experiments.The experimental results show that the MAE of SRAFE on E3 C benchmark is 0.666 and the PCC is0.809,which is greatly improved compared with the evaluation algorithm based on hand-crafted aesthetic features.It not only fully extracts the aesthetic information,but also introduces the aesthetic prior knowledge,which has strong interpretability and better performance of intelligent calligraphy evaluation.(4)For the practical application in calligraphy education,this paper designs the application of intelligent calligraphy aesthetics evaluation method in the form of APP,which is realized from the client side,server side and algorithm side respectively.It allows users to practice calligraphy copying synchronously with teaching materials,and generate objective and accurate intelligent aesthetics scores and intuitive visual guidance images.