Feature preserving milli-scaling of large format visualizations

Ultra-scale data analysis has created many new challenges for visualization. For example, in climate research with two-dimensional time-varying data, scientists find it crucial to study the hidden temporal relationships from a set of large scale images, whose resolutions are much higher than that of general computer monitors. When scientists can only visualize a small portion (< 1/1000) of a time step at one time, it is extremely challenging to analyze the temporal features from multiple time steps. As this problem cannot be simply solved with interaction or display technologies, this paper presents a milli-scaling approach by designing downscaling algorithms with significant ratios. Our approach can produce readable-sized images of multiple ultra-scale visualizations, while preserving important data features and temporal relationships. Using the climate visualization as the testing application, we demonstrate that our approach provides a new tool for users to effectively make sense of multiple, large-format visualizations.