Calligraphy Characters Generation With Specific Style Based On Calligraphy Knowledge

Calligraphy is an important part of China's excellent traditional culture.The development of calligraphy is of great significance to the development of Chinese excellent culture.In the long years of calligraphy evolution,many outstanding calligraphers have appeared and many wonderful calligraphy works have been left.However,due to the age,the calligraphy characters preserved in these works are very limited,making it difficult to learn and use.How to use a small amount of calligraphy characters to generate a complete calligraphy font library with specific style has become a new research hotspot.In the existing generation method,in order to realize the constraint of the Chinese character structure,it generally includes two steps of stroke extraction and stroke transformation.Stroke extraction generally uses the Coherent Point Drift(CPD)to match the skeleton of a given calligraphy character.However,this method can only obtain the skeleton point of the stroke and cannot deal with the situation of complex glyphs.Stroke transformation generally uses a neural network or a depth model to establish the transformation relationship,but this method can also only obtain skeleton points,and the method of using the rendering model to further obtain Chinese character glyphs is also likely to cause distortion of the glyphs.Calligraphy has complicated shape,and existing methods are difficult to solve the problem of the generation of calligraphy characters of a specific style.In view of the above problems,this article first uses image semantic segmentation and calligraphy character recognition technology to semantically segment calligraphic fonts.According to stroke categories we can obtain calligraphic font glyph points.And transform traditional two-dimensional skeleton point matching problems into threedimensional to avoid matching errors caused by crossed strokes.In addition,the semantic segmentation of the strokes can obtain the semantic information of the outline of the glyphs.Based on this,a complete stroke image can be obtained,and the calligraphy strokes can be extracted.Secondly,according to the knowledge in the field of calligraphy,this paper builds a stroke model.We use the ‘kaiti' font as the reference font,and proposes an automatic method for extracting the strokes of calligraphy characters based on the reference fonts,to achieve the automatic extraction of the strokes of calligraphy characters.Finally,this article uses deep learning technology to establish a conversion model from ‘kaiti' Chinese character strokes to specific style calligraphic character strokes,to achieve the conversion of ‘kaiti' Chinese character strokes to specific style calligraphic character strokes.And use the converted strokes to draw complete calligraphic characters.In order to further improve the generation quality,this paper establishes a rendering network to eliminate the rough edges and local distortion of the generated calligraphy characters.The experimental results show that the application of stroke semantic segmentation can effectively deal with the problem of skeleton matching and stroke extraction of complex glyphs;the application of the stroke model decomposes the learning process of the deep model,reduces the difficulty of learning,and effectively cope with the distortion of calligraphy glyphs in the deep learning method problem.