Color Style Transfer Of Map Area Symbol Based On Generative Adversarial Networks

With the application and development of maps in many fields,more and more people are involved in the making of maps.Increasingly complex social needs lead to higher requirements for maps,the use of maps and forms of expression are more flexible and novel.Under the promotion of Internet technology,the transmission and sharing of map works are more and more convenient.Excellent maps are more likely to be recognized and circulated by the public,how to use excellent map works of color collocation for their own use,become the goal of every cartographer.Therefore,more and more scholars pay attention to automatic color matching and style transfer in order to solve the difficulty of color selection for cartographers.In traditional cartography,cartographers mainly transfer map styles by hand,but the process is cumbersome and inefficient.In recent years,computer technology and deep learning promote the development of graphics and image technology,face recognition conversion and image style transfer methods are becoming mature.There are many similarities between maps and images,therefore,inspired by image style transfer,we introduce generative adversarial network(GAN)into map area symbol color style transfer.Which solves the problems of difficult color selection,long time consuming and low efficiency in cartography.The main research results are as follows:(1)The method of map color matching is analyzed,and excellent maps are collected.The style transfer of these maps is carried out based on Pix2 Pix model.In the process of style transfer,the model learns the rules of map color transfer through training.Realized the different hue,different saturation and different brightness color style automatic transfer.In the thematic map represented by zonal statistical method,the non-area symbol colors will influence the style transfer of area symbol colors,which will lead to the coloring disorder of the transfer result.In order to solve this problem,the interference color elimination algorithm is introduced.Which optimizes the result of area symbol color style transfer and achieves the transfer of area symbol color in the process of map style transfer.(2)Using Pix2 Pix to transfer the style of only one map is not ideal.In addition,it can only train sample maps of the same style at one time,and all the resulting maps will have the same style.The transfer of multi-style training set to multi-style results cannot be implemented.Therefore,Cycle GAN is used to experiment the style transfer of single multi-style map.Experimental results show that Cycle GAN can better complete the style transfer of single map.After only one training,different styles of maps can be generated according to the different styles of map samples.Compared with Pix2 Pix,Cycle GAN’s overall map style migration is better,but local style is easy to be lost.In the face of maps with fewer elements,there is a big gap between the migration result and the expected result.