Fact · Speech recognition
For OCR evaluation, character error rate is based on Levenshtein distance: the minimum number of character substitutions, deletions and insertions needed to turn the ground truth text into the OCR output, divided by the number of characters in the ground truth.
Checked against its source on October 17, 2026
In the source's own words
“It is based on the concept of Levenshtein distance, which measures the minimum number of character-level operations (substitutions, deletions, and insertions) necessary to transform the ground truth text into the output generated by OCR.”
Zero-shot OCR Accuracy of Low-Resourced Languages: A Comparative Analysis on Sinhala and Tamil, Nevidu Jayatilleke, arXiv, 2025, Evaluation Mechanism.
Open the original- What it measures
- The study defines CER and WER for scoring OCR output against reference text, using the same edit-distance idea at the character and at the word level.
- What it does not show
- The definition is standard in OCR research. The study does not compare the two measures on the same output.