A Semantic Recognition Based Automatic Color Palettes Design Tool

Computer-aided art design(CAAD)such as computer painting,computer automatic coloring,etc.has long been a domain that researchers explored.Because it is related to the diversity of human expression,making it much so challenging.With the development of artificial intelligence and machine learning in recent years,there has been some breakthroughs in CAAD.For example,GAN can automatically produce pictures,pixel2pixel can automatically fill colors according to existing styles,and PaintsChainer can let the secondary meta picture is automatically colored.This paper designs and implements an automatic coloring system based on semantic intelli-gence understanding algorithms and models.The issues mainly solved is the current automatic coloring system can only color by labels or relying on pre-colored images.This thesis proposed system is more intelligent and responding directly to human language.For designers,this system can instantly convert the descriptions of the texts into color palettes and color the sketch,which greatly facilitates its convenience.The color palettes comes from artworks and has a unique aesthetic that can help designers.This paper proposes three innovations to achieve this goal:1.Design and implementing a deep learning model to understand the semantics;2.Correlating picture information with hidden information at the semantic level.The partiality characteristics of the picture style similarity are preserved to its partial description of semantic description;3.In this paper,Flood Fill Algorithm and Random Walk are used to automatically fill in the image.There are three difficulties exists in this thesis:1.Representation and Parsing:Using word2vec to semantically understand the texts and artworks of the input texts,and designing the semantic information.2.Semantic Retrieval:Based on the information of the input texts and the informa-tion of the artworks,search for the artwork with the closest distance;3.Auto-Color:Extracting the color features of artworks,and keep the color continuity,color ratio,and light and shadow effects of the effect map.This algorithm has achieved good results in the actual testing of interior design linear draw-ings,and has achieved the best results in the industry.