Chinese Character Glyph Aesthetic Score Based On Deep Learning

People can’t avoid dealing with Chinese characters in daily life,and the research on aesthetic evaluation of Chinese characters writing is of great practical value,both from the perspective of calligraphy culture inheritance and people’s Chinese character teaching practice.At present,there are many ways to evaluate the aesthetic feeling of Chinese characters,and the combination of evaluation criteria and computer technology is still under study.In this paper,Chinese character font information was extracted from Chinese character font images by convolution neural network,and the aesthetic feeling of Chinese characters was studied from the external factor of font constraint and the internal factor of Chinese character stroke information.According to the aesthetic feeling research,the aesthetic feeling scoring system of handwritten Chinese characters was designed based on three characteristic information of font convex hull center,physical center and feature map.Therefore,the main work of this paper is as follows:1.In view of the inaccuracy of obtaining Chinese character font information after extracting HOG features from Chinese character font images,an algorithm for obtaining Chinese character font information based on VGG16 network was proposed.Firstly,the Chinese character glyph image was preprocessed,and then the convolution feature map was extracted from the preprocessed Chinese character glyph image by using VGG16 network.Secondly,the convolution feature map was used to obtain Chinese character glyph information,which was processed as the final Chinese character glyph information.100 Chinese character glyph images in Tiny Mind Calligraphy Chinese Character Recognition Competition were selected for experiments.The experimental results showed that the Chinese character glyph information obtained by this algorithm is more accurate,and the general rules would be more rigorous when applied to the research between different types of Chinese character glyph constraints and Chinese character aesthetic feeling.2.In view of the difficulty in writing beautiful Chinese characters at the beginning,this paper studied the relationship between the constraint mode of Chinese characters,which was an external factor,and the aesthetic feeling of Chinese characters.Therefore,this paper designed a handwritten Chinese character aesthetic scoring system based on three feature vectors.Firstly,the font image was preprocessed,and then different types of people are selected for Chinese character visual aesthetic experiment,which made the obtained aesthetic evaluation data more full and comprehensive.Secondly,the VGG16 network was used to extract Chinese character font information from the experimental Chinese character font image.Finally,the diagonal length,aspect ratio and four kinds of center eccentricity of Chinese character font were calculated.The aesthetic evaluation data of all Chinese characters were analyzed,and the results showed that there are differences in height,fatness and center position when people thought Chinese characters were beautiful under different Chinese character font constraints.Mastering these general rules under different Chinese character font constraints would enable people to write beautiful Chinese characters easily and quickly.3.In view of the fact that the external factor of font constraint could not substantially improve people’s writing of beautiful Chinese characters,this paper put forward an all-round projection measurement method to study the relationship between the internal factor of horizontal stroke information,left stroke information,vertical stroke information and the aesthetic feeling of Chinese characters.Firstly,Chinese character glyph images of various fonts were selected to form experimental data sets.Then,the convolution feature map was extracted from all Chinese character glyph images,and the convex hull center of Chinese character glyph was obtained according to the convolution feature map.Secondly,the whole-circle projection of the convolution feature map was performed with the center of the convex hull as the center of rotation,so as to obtain the stroke information of horizontal,left,vertical and pressing in each Chinese character glyph image.At last,the average value of the whole circumference projection metric values of all Chinese character font images in each type of font was taken as the whole circumference projection metric value of the Chinese character font.Mastering the writing skills of strokes in each Chinese character font can improve people’s ability to write beautiful fonts in essence,and also enable people to appreciate the cultural charm of different historical dynasties from the twists and turns of strokes.4.In the previous study on the aesthetics of Chinese characters,it was found that the physical center feature vector of Chinese characters,the central feature vector of the convex hull of Chinese characters,and the nineteenth feature map vector of the first-level convolution feature map play an important role in the aesthetic score of Chinese characters.Firstly,three feature vectors of the standard sample set of glyphs and the images of the sample set to be tested are obtained.Calculated cosine values of each feature vector of font images in the sample set to be tested and feature vectors corresponding to font images of the same Chinese character in the standard sample set,and took the average value of cosine values corresponding to the same score as the cosine value of this score.Then calculated probability values corresponding to different scores by softmax function,screened out the larger three cosine values and got the mathematical expectation of the corresponding score.Took it as the aesthetic evaluation value of the image to be tested and compared it with the real evaluation value.The experimental results showed that the handwritten Chinese character aesthetic scoring system can accurately score people’s handwritten Chinese characters,and could quickly improve people’s ability to write beautiful Chinese characters.