Binary AlphaDigits: 20x16 Binary Images of Digits and Letters

39 examples each of handwritten digits '0' through '9' and capital letters 'A' through 'Z'. The dataset consists of binary 20x16 pixel images, originally sourced from Sam Roweis and shared under a CC-BY-4.0 license on OpenML.

Use Cases

• Train a binary image classifier to recognize digits based on the 10 digit classes.
• Train a multi-class classifier to distinguish between all 36 alphanumeric characters.
• Benchmark feature extraction methods for low-resolution, binary character images.
• Study the separability of handwritten letter and digit classes in a simple binary image format.

Strengths

• Provides a balanced set with 39 examples for each of the 36 character classes.
• Uses a simple, well-defined binary image format (20x16 pixels) suitable for foundational experiments.
• Released under a permissive CC-BY-4.0 license for open use and redistribution.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for larger-scale training.
• Last update date is unknown; freshness unverified.

Provenance

Source

Sam Roweis

Collection Method

Likely collected from handwritten samples and converted to binary images.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null