Angle-Uniform Parallel Coordinates

In today’s data-driven era,datasets from various application areas have become increasingly complex,exhibiting large-scale and multi-dimensional features.This complexity poses challenges for information extraction and analysis.As a result,improving the efficiency of expressing and analyzing the information contained in data has become a critical issue in the field of visualization.Currently,parallel coordinates is one of the most widely used multidimensional visualization techniques and has unique advantages in assisting users with visual analysis and obtaining data insights.These advantages include the ability to plot multiple dimensions of data on the same plane,track changes of a single data point across multiple dimension axes,analyze the correlation between different dimensions using moving dimension axes,and easily incorporate new data dimensions by extending a new coordinate axis in parallel coordinates.However,traditional parallel coordinates suffer from some inherent flaws.Firstly,the mapping of correlations is non-linear and asymmetric,which hinders quantitative analysis by users.Secondly,visualizing large-scale datasets leads to a large number of overlapping data lines,resulting in cluttered parallel coordinates that interfere with user observation and thinking.These limitations restrict the application scenarios of traditional parallel coordinates and make it more challenging for users to explore data patterns.Therefore,proposing a new parallel coordinates model to enhance the visual analysis capability of large-scale multidimensional datasets is a highly challenging problem.In this thesis,we introduce angle-uniform parallel coordinates,a novel approach that deforms the image plane of traditional parallel coordinates to achieve a linear and symmetric mapping of data correlations.This approach enables efficient interactive analysis of large-scale datasets.Firstly,we summarize six design considerations of the ideal parallel coordinates and propose a new mapping model based on these considerations to achieve a linear and symmetric visualization effect,allowing for the display of both positive and negative correlations in data.Secondly,to address the visual clutter problem caused by large-scale datasets,we propose a combined subsampling and density visualization approach for angle-uniform parallel coordinates.This approach can provide consistent and clear visualization of global and important local patterns in the original data,while preserving outliers.To leverage interactive tools in the complete data analysis pipeline,we propose two new interactive forms:corner filtering and density plot brushing.These intuitive tools can highlight significant data features in the visualization results,facilitating user exploration and tracking of specific data subsets of interest.Finally,we validate the reliability and effectiveness of our method using synthetic and real-world datasets.Through case studies and comparative discussions,we demonstrate that our method has unique advantages in revealing data correlations and handling large-scale multidimensional datasets.In summary,this thesis aims to address the inherent problems of non-linear and asymmetric correlation mapping in traditional parallel coordinates,enabling quantitative analysis based on parallel coordinates.Our visualization approach designed for large-scale datasets can effectively reduce visual clutter and display data patterns of different scales in the dataset,while supporting interactive exploration and analysis by users.Finally,the comparison experiment results of large-scale datasets demonstrate that our method outperforms commonly used traditional methods with significant improvements in the readability of visualization results and the effectiveness of revealing correlations.