Optical character recognition (OCR), or text recognition, is a technology that automatically extracts text from images and converts it into a machine-readable format. For example, an OCR program can quickly convert JPG to Word, pulling data from scanned documents, camera images, and image-only PDFs.
The software identifies letters in an image, assembles them into words and sentences, and allows the original content to be accessed and edited. This process eliminates the need for manual data entry.
OCR systems use a combination of hardware and software to turn physical documents into machine-readable text. Hardware, like an optical scanner, copies or reads the text, while the software handles the advanced processing.
Modern OCR software often uses artificial intelligence (AI) to perform more advanced methods, such as intelligent character recognition (ICR), to identify different languages or handwriting. Organizations commonly use OCR to convert printed legal or historical documents into searchable PDFs that can be edited like a word processor file.
How does OCR work?
OCR software uses a scanner to process a physical document or an existing image, such as when you need to convert JPG to Word, and transforms it into editable, digital text. OCR can function as a standalone program, an application programming interface (API), or a web-based service. The process generally involves these steps:
- Image acquisition: All document pages are scanned, and the OCR engine converts the digital document into a black-and-white version. The software analyzes the image, identifying dark areas as characters to be recognized and light areas as the background.
- Preprocessing: The digital image is cleaned up to remove unwanted pixels. This step can include deskewing to correct alignment issues, removing graphical elements like boxes or lines, and identifying if script text is present.
- Text recognition: The dark portions are processed to find letters, numbers, or symbols. This stage typically focuses on one character, word, or block of text at a time. Characters are identified using one of two algorithms: pattern recognition or feature recognition:
- Pattern recognition: The OCR program compares characters in the scanned document to a database of stored text examples (glyphs) in various fonts and formats. For this to work, the font must already be in the OCR’s database. Given the vast number of fonts and languages (like Arabic, Chinese, English, French, and Spanish), training a system on every combination would be resource-intensive.
- Feature recognition: This method is used when the OCR program encounters a font it hasn’t been trained on. It applies rules based on the features of a character, such as the number of angled lines, intersections, or loops. For instance, the letter “A” is identified as two diagonal lines connected by a horizontal line. Once identified, the character is converted into an ASCII code that computer systems can manipulate.
OCR technology
- Layout recognition: A more advanced OCR program analyzes the document’s structure, dividing the page into elements like text blocks, tables, or images. Lines are broken into words and then characters. After isolating characters, the program compares them to pattern images and, after processing all likely matches, presents the recognized text.
- Postprocessing: The extracted information is saved as a digital file, either in an editable format or as a PDF. Some systems save both the original image and the OCR version for easier comparison and document management.
Types of OCR
OCR programs come in four main types, each getting smarter as you go:
- Simple OCR: This is the most basic type. It matches scanned characters to stored ones, like comparing patterns. It’s not super flexible because there are so many fonts and languages out there.
- Optical Mark Recognition (OMR): Think checkboxes, survey bubbles, signatures, and logos. OMR scans for these marks and matches them to stored images, similar to how simple OCR works.
- Intelligent Character Recognition (ICR): This one’s a step up — it uses AI! The program learns as it goes, using machine learning to get better at reading by practicing. It focuses on curves, lines, and intersections to figure things out.
- Intelligent Word Recognition: This takes ICR to the next level. Instead of focusing on single characters, it recognizes whole words at once, making it much faster.