Optical Character Recognition (OCR) Systems process scanned text into text usable by computers. We observe that different OCRs make independent mistakes. This example uses a simple Logistic Regression encoded in our system, to select between OCR outputs when they differ.

This example uses outputs from two open-source OCRs for a dataset of 620 words, whose features are already extracted. The dataset is hand- labeled.

• PostreSQL
• Python
• Matplotlib (pip install matplotlib)

• If necessary, modify db.url to fill in your database connection details.
• Execute deepdive initdb to initialize database.
• Execute deepdive run.

• Execute python feature-analysis.py.
• Feature analysis and system calibration result are in output/ and run/LATEST/calibration/, respectively.
• For details, run deepdive sql to examine the result relations.