Traditional Pattern Generation Algorithm Research And System Implementation Based On Global And Local Stylization Constraints

In the era of digital intelligence,how to propose a set of image generation algorithm with independent intellectual property rights based on the Chinese cultural material database,and provide materials and tools for the overall digitalization of cultural creation and production has become an urgent problem.Based on the Chinese cultural material database,this paper puts forward two algorithms for the generation of Chinese traditional patterns,which are satisfied with the original generation and stylized generation of Chinese traditional patterns,and used for the storage and supplement of Chinese traditional pattern material database.The research content mainly includes:(1)In view of the fact that the conventional generative adversarial network can not be effectively applied to the generation of the traditional pattern authenticity,this paper proposes a sequential learning pattern image formation method,and through the pyramid level control pattern authenticity of a space sensitive adversarial network,solves the storage problem of the traditional national pattern digital process.Experiments show that the method proposed in this paper can effectively store the original pattern,and the pattern regenerated by the model has achieved excellent results both in the overall structure and in the local details.In sifid and is and other objective indicators,they are better than the traditional pattern generation methods.At the same time,the proposed method can efficiently generate any resolution pattern image,and be applied to the task of traditional pattern super-resolution.(2)In the study of traditional pattern generation,parametric method and non-parametric method both lose the image style information to some extent.This paper improves the results of traditional pattern generation from two aspects.On the one hand,the high-dimensional features of the image extracted by the pre-training convolutional neural network are normalized in different spatial representations,which enhances the fusion of low-level fine spatial information and high-level semantic information;on the other hand,global and local style constraints are integrated to control the coordination of global and local style information.Experimental results show that the proposed method can effectively enhance the smoothness of the generated results and enrich the image style.In addition,the objective indexes such as consistency,smoothness and aesthetic evaluation score are comprehensively evaluated.(3)A generation system of traditional patterns is designed and implemented.The system can support the original generation and stylized generation of traditional patterns.This paper puts forward two methods of original generation and stylized generation of traditional patterns to support the creation of cultural materials and provide design tools.