Research On Content And Style Recognition Of Calligraphy Characters Based On Deep Learning

Calligraphy has its unique artistic beauty and is one important part of the traditional Chinese culture.With the advancement of image digitisation technology,calligraphy recognition research is beginning to be valued by scholars and has important research implications in image retrieval,calligraphy content recognition,calligraphy style appreciation,calligraphy teaching and evaluation,and calligraphy annotation.However,the scarcity of calligraphic works,the interference of noise during circulation,and the difference of calligraphic styles make it more difficult for the traditional optical character recognition technology and some early feature extraction technology.With the continuous development of artificial intelligence technology and the promising progress in the field of image recognition,new research directions have emerged in calligraphy recognition.Applying deep learning techniques to the field of calligraphy recognition can address the limitations of traditional feature extraction.However,the current calligraphy recognition has problems such as single research perspective and low recognition accuracy.In order to broaden the research field of calligraphy recognition and further improve the accuracy of calligraphy recognition,this paper proposes a method for synchronous recognition of calligraphy character content and style based on deep learning and Label Power Set(LP)method,and a method for synchronous recognition based on deep learning and transfer learning.In order to solve the problem of single perspective of calligraphy recognition and single label of the existing calligraphy image dataset,this paper creates a new calligraphy character image dataset,and proposes a synchronous recognition method of calligraphy character content and style based on deep learning and LP method.Firstly,the training samples of calligraphy images are processed using the LP method to label the combination of calligraphy character styles and contents as a label category.Secondly,a convolutional neural network for synchronous recognition of calligraphy content and style is constructed by using Py Torch deep learning framework and residual network structure.Finally,the network is trained to convergence according to the training set processed by the LP method,and the performance is tested based on the dataset constructed in this paper.It is experimentally verified that the recognition method based on deep learning and LP method proposed in this paper achieves a large improvement in accuracy compared with the traditional calligraphy recognition research method.In order to improve the accuracy of synchronous recognition of calligraphic character content and style,this paper proposes a synchronous recognition method based on deep learning and transfer learning.Firstly,considering the existence of connection between calligraphy styles,this paper builds a calligraphic style recognition network using the characteristic of structural feature reuse in dense networks,trains on the training set until the network converges and conducts performance tests.Secondly,the parameter weights for the convergence of the recognition network based on the LP method are obtained using the model parameter transfer technique in transfer learning,and the calligraphic content recognition model is initialized with these parameters,retrained on the training set until convergence and tested for performance.Finally,the calligraphy style network is combined with the content network to complete the synchronous recognition of the content and style of calligraphy characters.It is experimentally verified that the synchronous recognition method based on deep learning and transfer learning proposed in this paper has higher recognition accuracy,and also has more application value than the traditional research of single branching task.