Automated Colormap Generation For Categorical Data Visualization

Categorical data is a type of data used to describe categorical attributes.It is commonly used to represent discrete variables that do not have quantitative or ordinal relationships and is a common data type in abstract information spaces.To effectively represent the information contained in categorical data,people use visual encoding elements such as color,shape,transparency,etc.,to represent the abstract data as intuitive graphic symbols.Among the various visual encoding elements,the application of color is the most widespread.By establishing a mapping relationship between color and data,we can generate visually rich visualizations.This not only accurately represents data category distributions and semantic information but also attracts and maintains the reader’s attention by adjusting color contrast differences.The design of a color mapping scheme consists of two stages:(1)palette generation,which involves creating a palette containing several colors;and(2)color assignment,which involves assigning colors from the palette to different categories of data.In practical applications,users often use existing color design tools such as Tableau and ColorBrewer to select system-default palettes or design palettes automatically based on empirical color theories.They then further manually adjust colors to assign them to different categories of data.Users typically need to spend a significant amount of effort in a trial-and-error process,to achieve discriminable visualization results.When there is overlap between different categories,people often set different levels of transparency for each category to achieve a semi-transparent rendering of all the distributions for better discrimination.However,people perceive different colors differently in semi-transparent scenes,and some colors may produce new colors when blended semi-transparently,leading to ambiguity in understanding category information.Furthermore,the existing automatic design algorithms have not taken into account the impact of color modifications on category information during interactive exploration,resulting in insufficient distinguishability in coloring results and poor performance across different interactive analysis tasks.Therefore,generating highly discriminative color mappings suitable for different visualization application scenarios based on given data is an urgent problem in data visualization.To address the above issues,we explore the automatic generation of color mappings for categorical data in three different scenarios:opaque coloring and translucent coloring in static visualization,as well as interactive highlighting coloring in dynamic visualization.The main contributions of this dissertation are as follows:(1)Discriminable Colorization for Categorical DataIn order to reduce the trial-and-error cost for users in generating visualization results and enhance the efficiency of conveying information through visualization,this dissertation proposes an integrated approach-Palettailor,for creating and assigning color palettes to different visualizations.This approach takes data characteristics into account to produce color palettes,which are then assigned in a way that fosters better visual discrimination of classes.To do so,Paletailor uses a customized optimization based on simulated annealing to maximize the combination of three carefully designed color scoring functions:point distinctness,name difference,and color discrimination.The efficiency of the optimization allows users to incorporate modifications into the color selection process.This method supports automatic colorization for various visualization types,including scatterplots,line charts,and bar charts.(2)Optimizing Color Assignment and Opacity for Reducing Misinterpretation of Overlapping VisualizationsIn order to reduce color confusion and enhance cognitive associations between different colors,this dissertation proposes an automated approach for generating optimal color assignments and opacity values for overlapping visualizations,with extension to multiple visualizations that employ transparency to improve the accuracy in perceiving the underlying distributions,such as overlapping histograms,parallel coordinates,and cluster-encapsulating hulls.This approach exploits color name information to ensure that the resulting colors are semantically connected to the original class labels,while also maintaining perceptual discriminability and separability between the distributions.By a customized simulated annealing algorithm,this method can rapidly generate optimized color mappings,while ensuring solution diversity.(3)Interactive Context-Preserving Color Highlighting for Multiclass ScatterplotsIn order to enhance the efficiency in acquiring context information during interactive highlighting,this dissertation proposes a context-preserving color highlighting method for the interactive exploration of multi-class scatterplots,to achieve desired pop-out effects while maintaining good perceptual separability among all classes and consistent color mapping schemes under varying points of interest.This approach simultaneously generates two contrastive color mapping schemes with large and small contrasts to the background,while maintaining good perceptual separability among all classes and ensuring that when colors from the two palettes are assigned to the same class,they have a high color consistency in color names.Then interactively combine these two schemes to create a dynamic color mapping for highlighting different points of interest.This method can also support other visualization types,such as line charts and bar charts.