Synthetic IRS Tax Form Images With Hand-Printed Characters

900 simulated tax submissions contain 20 different 1988 IRS form faces, averaging 6.22 forms per submission. The National Institute of Standards and Technology created this binary image database for testing form recognition systems, using computer-derived images that mimic real hand-printed forms but contain no actual taxpayer data.

Use Cases

• Train and evaluate forms recognition software on 20 distinct IRS form faces like 1040 and Schedules A-F
• Benchmark hand writing recognition algorithms on simulated hand-printed characters from tax documents
• Test optical character recognition (OCR) systems on binary images of structured forms with ground truth
• Develop software for forms identifications across a set of 12 different tax forms and their multiple pages

Strengths

• Contains 900 complete simulated tax submissions for realistic testing volume
• Includes 20 different form faces representing 12 distinct IRS forms and schedules
• Provides ground truth for forms and character recognition without real taxpayer data
• Created by the National Institute of Standards and Technology for authoritative benchmarking

Limitations

• Data is synthetic and computer-derived, which may not fully capture real-world variation and noise
• Limited to 1988 IRS tax forms, making the form layouts and content historically outdated
• No columnar or structured data is provided, only binary images and associated ground truth

Provenance

Source

National Institute of Standards and Technology

Collection Method

Images automatically derived and synthesized using a computer to mimic real hand-printed forms.

Time Range

Forms are from the 1988 tax year.

Freshness

Last updated in June 2023, though the underlying data represents 1988 forms.

Geography

United States (IRS forms).