Stoke Segmentation And Application Of Calligraphy Combining VDSR And Graph-cut

Chinese calligraphy has a long history in China,is a unique Oriental art form,which contains a variety of writing skills and aesthetic ideas,is a unique cultural treasure of China.In order to make calligraphy characters move forward from traditional rice paper to the development of computer information,it is an indispensable step for calligraphy learning to go digital to study the vectorization method of calligraphy Chinese characters strokes,which aims at dismantling strokes.As the stroke types of Chinese characters in calligraphy are very diverse,and there are many different fonts,such as regular script,running script,and running script,the types of Chinese characters in calligraphy are many times more than the number of commonly used Chinese characters,all of which pose a challenge to the generalization of the method of disassembling Chinese characters.Many existing methods of disassembling strokes often encounter ambiguity in the cross area.In addition,calligraphy and Chinese characters have heavy strokes,and the outline of strokes is formed by special writing skills,so the outline of strokes is hidden in the overlapping part.How to restore the missing outline information is another difficulty in disassembling strokes.Therefore,the research on the stroke segmentation method of Chinese calligraphy becomes more realistic.This paper analyzes the defects of the existing semantic vectorization methods,proposes many effective improvement methods,and applies the method to the practice of calligraphy intelligent education.The main contents are as follows:(1)a convolutional network structure is proposed to calculate the complete path correlation.This paper improves the VNet method,and points out the problems in the process of training path correlation.Secondly,the path correlation should be binary,not related to the gray value of the pixel.Finally,considering that the problem requires the convolution network to have a large range of receptive field,we adjusted the network depth and the size of the convolution kernel of the VDSR model to obtain a complete receptive field,and the final prediction result obtained a higher intersection ratio.(2)a method is proposed to predict the overlapping degree of cross regions based on the global correlation elevation map.In this paper,the concept of globally correlated elevation map is put forward,which plays a key role in restoring the potential contour of overlapping areas and provides a new idea for solving the contour loss caused by overlapping in stroke segmentation field.In the IOU comparison of predicted results,our method is better than other methods.The existing overlap region prediction method solves a classification problem,which is generalized as a regression problem,and based on the predicted overlap amount,the structure of the graph is augmented by multiple nodes,and the multi-overlap problem is finally solved.(3)in this paper,the methods of vector Chinese characters and calligraphy online intelligent education practice,research and development of the calligraphy online intelligent education platform,on the one hand,enabling users to stroke outline of calligraphy and dismantling,strengthen the deep going study,on the other hand,Chinese characters stroke vector quantization method for the evaluation method of Chinese characters,Chinese characters beautification method,Chinese character style migration method provides technical precursors such as digital calligraphy field.