Optical Character Recognition
Also known as OCR
Overview
Optical character recognition (OCR) studies how to detect and transcribe text in images, enabling search, accessibility, digitization, and document processing. Common approaches include detection-and-recognition pipelines with CNNs, recurrent networks, and CTC decoding; Transformer recognizers; and end-to-end vision-language models, often adapted with synthetic data or language-specific fine-tuning. Evaluation spans scene text, forms, historical manuscripts, mathematical expressions, and multilingual documents, using measures such as character error rate alongside layout and downstream-task benchmarks. Large vision-language models handle varied document tasks, but specialized smaller systems can be faster or more accurate in particular settings, while low-resource scripts, degraded images, and fluent but visually unfaithful corrections remain difficult.
Latest papers
205 papers21 papers in the last four weeks, down 9% on the four weeks before. 0.1% of all new papers.