The Research Of Graphic Image Generation Based On Design Features

With the development of technologies such as big data,machine learning,and human-computer interaction,the process of design is moving towards a new form.Using algorithms to help designers complete repetitive design tasks is one of the important contents of intelligent design research.Graphic design is a very important design form in design tasks.It is a challenge to solve large numbers of design requirements which contain low-value design contents,such as posters and various type of advertisements.These designs are always discarded when used up,which not only wastes the social resources,but also wastes the inherent value of design resources.This thesis takes the advertising image as the target,and proposes a method for advertising image generation based on the modeling of graphical design features.By studying the design rules of graphic design,a hierarchical feature model is proposed.Our method establishes several probabilistic models by measuring the geometric features,perceptual features and style in the image.These models could fit discrete data samples to estimate the probability density distribution of features under the target conditions.Based on the optimal features obtained by sampling,our method further generates the layout and recolors the image,which thus obtains the result images with diverse forms.Specifically,the main content of this thesis can be summarized as follows:1.A hierarchical feature model for advertising images is proposed.According to the geometric features,perceptual features,and image style in the hierarchical model,the methods for measuring features are proposed.The contents of the study include the method of color extraction,the calculation of perceptual features,and the pairwise comparison method of image style.2.A method for modeling design features based on probability estimation is proposed.Aiming at the layout and color of graphic design elements,the model of probability density estimation is studied.The contents of the study include feature sampling method,model optimization method,and data learning strategies.3.A method of advertising image generation according to perceptual features is proposed.Based on layout clustering and perceptual feature visualization,the methods of text layout generation and element recoloring are studied.The contents of the study include the probability estimation of characters' layout,the optimization method based on perceptual features,the geometric mapping method of chromatic value,and the dynamic adjustment strategy of brightness value.Based on the proposed data,models and methods,this thesis develops a system of automatic ads generation,and verify the method and system by implementing the design task of clothing advertisements.The methods proposed in this thesis explore the implementation of the intelligent advertising image design,by detailing and elaborating related methods and researches.Studying the method of advertising image design based on feature modeling is not only of great significance for solving repetitive design tasks,but also provides a basis for the development of the intelligent design.