DRUID: Reconstructing Data from Scatter and Line Plots with An Automated Machine Learning and Computer Vision Pipeline*

Scientists use scatter charts and line charts to portray their data in publications and presentations. While some authors may include their datasets in their papers or published work, many do not, and this results in the reader having no way to access the data portrayed in these plots. Scientists may turn to artificial intelligence agents (AIs) to reconstruct the original data from the chart. Many freely available AIs and MLMs struggle with this task, requiring many more inputs or requests to extract the wanted information, and are expensive both in terms of resources and computational hours. DRUID, or Data Reconstruction Using Image Dissection, is our attempt at this issue. DRUID uses optical character recognition, straight line detection, image contour analysis, and other classical and artificial intelligence algorithms to reconstruct the original data from just an image of a line plot or a scatter plot.