FigTExT: A Tool for Extracting Text from Biomedical Figures
Author Information
Author(s): Kim Daehyun, Yu Hong
Primary Institution: University of Wisconsin-Milwaukee
Hypothesis
Can a specialized tool improve the extraction of text from biomedical figures compared to standard OCR tools?
Conclusion
The FigTExT tool significantly outperformed standard OCR tools in extracting text from biomedical figures.
Supporting Evidence
- FigTExT achieved 84% precision and 98% recall for text localization.
- FigTExT outperformed the standard OCR tool, which had only 36.6% precision.
- Using domain-specific lexicons improved the performance of text extraction.
Takeaway
FigTExT is a tool that helps scientists read text from images in research papers better than regular text readers.
Methodology
The study evaluated an OCR tool and developed FigTExT, which includes image preprocessing, character recognition, and text correction.
Potential Biases
Potential bias in the selection of figures and the evaluation of text importance by domain experts.
Limitations
Only 42.2% of figure texts appeared in the corresponding full-text articles, which limited recall.
Participant Demographics
Figures were selected from 70 full-text articles in PubMed Central.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website