MathCaptcha10K is a dataset of arithmetic CAPTCHA images designed to test and train CAPTCHA recognition models, especially when dealing with CAPTCHAs with distracting backgrounds and deformed text.

The dataset contains 10,000 labeled samples and 11,766 unlabeled document samples with a resolution of 200 × 70 pixels. Each labeled sample contains an arithmetic captcha image, the exact characters in the image, and its integer answer.