Intelligent Generation And Style Refinement Based On The Historical Excellent Layouts Of Digital Newspapers

Newspapers have always been an important carrier for people to stay informed and disseminate knowledge.In the traditional newspaper printing industry,designers need to manually typeset according to design rules,which is time-consuming and expensive.Currently,there is a lack of research on automated newspaper typesetting both domestically and internationally.To improve typesetting efficiency,this thesis proposes a method for automated layout generation and refinement based on excellent historical newspaper layouts.For a given news article,the first step is to infer the style of the electronic newspaper layout based on the probability model of historical excellent layout training,and combine fixed constraints and user constraints to ensure that the style is effective.At the same time,a quantitative method based on aesthetic design principles is constructed to further achieve style fine-tuning.In addition,due to the fact that the underlying layout system is a Latex layout system,in order to expand the Tex code representation of graphic and textual structures,this thesis introduces an automatic code generation framework based on Pix2 code,which is used to achieve automatic generation from news block images to Tex code.Finally,through qualitative and quantitative evaluation,we demonstrate that the new method can generate newspapers that meet visual aesthetics,hierarchy,and readability,and can effectively identify Tex codes corresponding to news block images,achieving the goal of enriching newspaper style templates.The main work and innovative achievements of this thesis are as follows:(1)This thesis contributes a fine-grained tagged dataset for electronic newspapers,which contains rich semantic information on design elements.This thesis also shows how to construct a mapping from images to tex code,in order to provide more samples and scenarios for the intelligent layout research of electronic newspapers.In addition,this thesis also contributes a news block image-code dataset,which contains the news block images and the specific codes corresponding to the images,to provide support for future research on the automatic generation model of tex code based on digital newspaper images.(2)This thesis proposes a method for automated layout generation and refinement based on excellent historical newspaper layouts.This method learns the design style of excellent historical newspaper layouts and achieves style refinement by combining user constraints and design principles to generate high-quality style designs.Our approach also demonstrates high effectiveness in typesetting result analysis and design comparison experiments,offering new insights and methods for automated typesetting in electronic newspapers.(3)This thesis proposes a framework for Tex code generation based on digital newspaper images.To address the problem of generating Tex code from digital newspaper images,this thesis improves and optimizes the Pix2 code model,mainly focusing on its visual model,language model,and decoding layer,using ResNeXt and GRU to optimize the network structure.Experiments demonstrate that the optimization of this model can achieve more accurate and efficient code generation,providing strong support for automated code generation in digital newspapers.