Zhao Nano Lab Releases LinePilot Digitizer: An Open-Source Tool for Scientific Plot Digitization
The Zhao Nano Lab at the University of Georgia is pleased to announce the public release of LinePilot Digitizer (LinePilot), a free, open-source software tool designed to recover numerical data from scientific line plots. LinePilot enables researchers to convert published figures, scanned plots, and other plot images into reusable numerical datasets, facilitating quantitative analysis, data comparison, and machine learning applications.
The project was led by graduate student Fengbo Ma, with significant contributions from undergraduate students Rayan Akhtar and Aakash Joshi, under the supervision of Professors Zhen Xiang, Haijian Sun, Xianyan Chen, and Yiping Zhao. The research team developed an integrated approach combining accurate axis calibration with continuous color-based curve extraction.
A key innovation of LinePilot is its integration of three complementary calibration modes:
Standard manual calibration: Allows users to define axis reference points directly.
Enhanced manual calibration: Automatically refines user-selected calibration points using nearby axis and tick features to improve accuracy.
Automatic OCR calibration: Uses optical character recognition to identify axis labels and estimate coordinate mappings with minimal user intervention.
All three modes share a continuous, color-based curve recovery algorithm that uses dynamic programming to trace plotted lines, providing flexibility in balancing automation, user control, and accuracy.
To systematically evaluate digitization performance, the team also developed DigitizerBench, a benchmark containing 2,980 synthetic plot images designed to test the effects of signal characteristics, image resolution, rendering conditions, and plot complexity.
Benchmark comparisons demonstrated that LinePilot achieved the best overall performance among the evaluated tools in both automatic and human-guided settings:
Automatic mode: LinePilot (OCR) achieved the lowest mean failure-penalized normalized error (FPC-NRMSE = 0.672) among the tested automatic methods.
Enhanced manual mode: LinePilot (enhanced) achieved a mean FPC-NRMSE of 0.081, with 100% output success and 93.3% trusted usability in the human-guided evaluation.
These results demonstrate LinePilot's potential as a practical tool for recovering quantitative information from published scientific figures.
LinePilot is freely available to researchers, educators, and students. The public release provides a desktop application, documentation, tutorials, and programmatic interfaces to support a range of scientific data extraction workflows.
Explore LinePilot
LinePilot represents another step in the Zhao Lab's broader efforts to develop open scientific software and research infrastructure. Complementing the lab's SpectraGuru platform, LinePilot helps bridge the gap between scientific information presented in published figures and reusable numerical data for computational analysis.
This research was supported by the National Science Foundation Pathways to Open-Source Ecosystems (POSE) Program (Award No. 2518273). The authors also acknowledge the contributions and support of the SpectraGuru development and stakeholder teams.
Comments