Deep Colormap Extraction From Visualizations.

Area of Science:

• Computer Vision
• Data Visualization
• Machine Learning

Background:

• Extracting colormaps from visualizations is crucial for data analysis and manipulation.
• Existing methods for colormap extraction lack accuracy and efficiency.

Purpose of the Study:

• To develop a novel deep learning approach for automatic colormap extraction from visualizations.
• To enhance the utility of visualizations through accurate colormap identification.

Main Methods:

• A deep neural network processes Lab color histograms of visualization images.
• An atrous spatial pyramid pooling module captures multi-scale color features.
• The network classifies colormaps as discrete or continuous and refines predictions.

Main Results:

• The proposed method achieves superior performance compared to existing approaches on synthetic and real-world data.
• A new dataset of approximately 64,000 visualizations was created for training.
• The approach demonstrated effectiveness in color transfer and remapping use cases.

Conclusions:

• Deep learning offers a powerful solution for automatic colormap extraction.
• The developed method provides accurate and efficient colormap identification.
• This work has practical implications for data visualization tools and workflows.