Research And Application Of Logo Image Intelligent Design From The Perspective Of Reception Aesthetics

Under the background of the era of artificial intelligence,there are more and more design scholars studying intelligent design,which includes digital media art,visual communication design,environmental art and product design,clothing wear design,etc.,as main directions,aiming to adapt to people’s intelligent lifestyle through intelligent design.Automated design can assist designers to complete repetitive and simple design tasks,and can replace part of manual design to improve efficiency and reduce time costs.Moreover,in the new era,internet users have a tremendous demand and consumption for graphic design works.Compared with the traditional design methods,the automatic design has faster update speed,and it can help users to quickly select and use design works,such as some application systems and related technologies that can achieve automatic generation of posters,the calculation of web page aesthetics,and the automatic generation of logos,etc.Among them,the logo is an essential part in the practice of graphic design,an aesthetic symbol to convey the visual culture of the brand,and a functional symbol to express the image information.Previously,many researchers have carried out the work of applying artificial intelligence technology to the intelligent design of images.However,there are still many deficiencies in AI technology-assisted design to be solved.It is mainly reflected in the following three aspects:(1)Artificial intelligence technology pays more attention to the improvement of technology,there is not much research in the field of design,the attention to the design-oriented issues is not high,and there are fewer people who study design issues in cross-research.(2)There are very few large-scale design material databases publicly available in the field of artificial intelligence.Special design material databases are mainly used to solve design problems.Special design ma-terial databases are very helpful for researching design problems.The ideas of technological improvement are different,so the lack of professional databases brings inconvenience to design-ers to participate in the cross research of artificial intelligence technology.(3)There is a lack of macro unified guidance,and there is a polarization between design-based and technology-based research,and research is biased towards design or technology,and it is difficult to balance multi-disciplinary integration.The main research work and contributions of this paper are as follows:(1)In the case that logos need to meet the requirements of both machine learning and aesthetic characteristics,this paper created a large-scale logo database that collected design materials: JN-Logo(JN-Logo Visual Analysis),which was used for image aesthetic evaluation and image intelligent design generation.Our database JN-Logo provides three types of annotations: aesthetic quality score,style attribute classification and semantic description,including 6 scoring standards,6 types of style labels and semantic descriptions.This paper established a benchmark for JN-Logo to measure the performance of algorithmic models that people utilize to select logo data.In order to prove its characteristics and effectiveness,the paper shows the advantages of JN-Logo database in five aspects,demonstrates the use and evaluation method of JN-Logo,and proves that high-quality logo data is more conducive to the application of graphic intelligent design.(2)In order to solve the time-consuming and labor-intensive problem that designers need to manually search for similar styles of logos and classify styles when researching and referenc-ing logos for the design,this paper proposes an automatic classification method that based on visual Transformer and deep clustering named DTCluster(Deep Transformer-Based Cluster)to analyze the styles of logo image data.In this section,several other typical clustering methods are comparatively analyzed with the method established in this paper,which proves that current method in this paper has a superior performance.At the same time,five types of semantic de-scriptions,overall harmony descriptions,and descriptions of colors,shapes,structures,styles,are proposed for the results of this method and other classical methods.This work provides a new semantic description for other researchers to study logo data.(3)In order to solve the time-consuming and labor-intensive problem that designers need to manually select multiple colors when design sketches,this paper proposes GAN network model that can logo images automatically.The method improved the previous method and can color the logo sketch innovatively.It only needs to provide one logo sketch to output multiple colorful logos with visual impact.Based on the pix2 pix network,this paper designed an image translation network based on contour generation.In order to make the image generation more stable,we improved the UNet module used inside pix2 pix,integrated the attention mechanism,and added a new loss function based on the pix2 pix objective function,which finally made the generator output pictures with more diversity and bright colors.(4)For users to realize their own design logo according to their own needs,this paper proposes a Cycle GAN network generation model that can generate specified style logo by in-putting sketches and style semantic descriptions.By introducing the semantic description of multi-style logo,the model automatically generates the specified style image from the sketch,and at the same time integrates the attention mechanism to improve the accuracy of the gener-ation.Finally,the generated image is further optimized by post-processing strategies such as binarized denoising.