Deep Learning Calligraphic Evaluation Modeling Incorporating Writing Movement Acquisition

Calligraphy is a well-deserved treasure in our traditional culture,combining practicality and artistry.With the development of computer technology in recent years.Digital calligraphy research has become an emerging field.However,the aesthetic evaluation of calligraphy in the field of digital calligraphy has been a hot topic of research and a difficult problem.Due to its lack of theoretical support and data support,aesthetic evaluation of calligraphy can only rely on comparing the similarity between calligraphic images to reach a calligraphic evaluation.And such an evaluation approach ignores traditional calligraphy theory,making aesthetic evaluation slow to develop.In an attempt to break through this bottleneck,this paper delves into the concept of writing movement in traditional calligraphy theory and finds that writing movement is feasible from a theoretical perspective to achieve calligraphic aesthetic evaluation.Based on this,this paper focuses on(1)the collection and quantification of data on writing movement in calligraphy.(2)the verification of the feasibility of writing movement data for evaluating calligraphy from a technical perspective,and the provision of an effective baseline model.The main research content and results of this paper are as follows:(1)By analyzing the mechanics of the writing movement in calligraphy,we designed the writing movement data recording device based on the nine-axis motion inertial sensor MPU9250.Subsequently,we designed a calligraphy copying experiment and used a writing movement recording device to collect writing movement data and pictures of calligraphy works from 50 people,and cleaned the data to form the first calligraphy writing movement data set.(2)With the support of calligraphy teachers,a modern calligraphy evaluation system was designed,and the dataset were evaluated aesthetically based on this system and the calligraphy writing videos.The evaluation label labeling work was completed for all the writing movement data and corresponding calligraphy pictures to provide data support for the next step of neural network training.(3)In order to verify the quality of the calligraphy pictures from the self-built dataset and to compare the evaluation effect with the evaluation model of the writing movement data for demonstration,this paper prioritizes the design of the aesthetic evaluation model for calligraphy pictures.After completing the pre-processing of calligraphy images,three convolutional neural network architectures were experimented for designing calligraphy evaluation models for calligraphy images.The Transfer Learning-based Residual Network was finally determined as the baseline model.(4)In order to verify the feasibility of calligraphy evaluation with writing movement data,Simple RNN network was firstly tried,and the content of dribbling data was determined based on the training effect of this model.Subsequently,Long Short-Term Memory network was used to design the calligraphy evaluation model for writing movement data,and finally the Long Short-Term Memory network combined with the K-Nearest Neighbor algorithm was determined as the baseline model for the dribbling data evaluation model.By randomly sampling the prediction results of different models,it was found that the aesthetic evaluation model based on calligraphic pictures lacked the ability to evaluate different grades of works with similar outlines but significantly different brush strokes,while the aesthetic evaluation model based on writing movement data was able to overcome this difficulty well and improve the accuracy by 7.1%.We also released a questionnaire for social research,the results of which showed that writing movement data can not only be an important reference factor in calligraphy evaluation,but also better screen minor problems than the common way of evaluating calligraphy works with pictures,and better assist beginners in their calligraphy learning.In summary,this paper proposes an aesthetic evaluation system for calligraphy with calligraphic writing movement data as the core.We collected relevant pen movement data through homemade devices and calligraphy copying experiments to create the first pen movement data set for regular script calligraphy.The subsequent work on the design of a calligraphic evaluation model for writing movement data was implemented based on the use of neural networks,providing new ideas for the study of aesthetic evaluation of calligraphy.