OCR

OCR, short for optical character recognition, is a technology that converts raster images of text—such as scanned documents, PDFs, or photographs—into machine-readable, editable text. It plays a key role in digitizing printed material and enabling searchable archives or automated data extraction.

Modern OCR workflows include stages such as image preprocessing to improve quality, layout analysis to identify text blocks, character segmentation and recognition using template matching, feature-based methods, or neural networks, and post-processing with spell checking and language models to reduce errors.

Historically, OCR emerged from pattern recognition research in the mid-20th century, with later advances driven by machine learning and deep learning. Practical systems distinguish machine-printed text from handwritten text, with handwriting recognition being more challenging. Popular OCR engines include open-source Tesseract and various commercial offerings, and some engines specialize in specific languages or scripts.

Applications include digitizing books and documents, converting receipts and invoices into structured data, enabling full-text search in document repositories, and assisting visually impaired users through screen readers and accessible formats.

Limitations include sensitivity to image quality, font variability, page layout complexity, and language or script. Complex layouts, noisy scans, or multilingual documents can reduce accuracy. Evaluation uses metrics such as character error rate and word error rate; high-quality scans of standard Latin scripts can reach high accuracy, while handwriting and exotic scripts remain more challenging.