What is OCR?

An OCR system analyses the structure of the document and divides it into various elements such as text blocks, tables and images. These structural elements are further broken down into lines, words and finally into individual letters. The letters are compared with a database of sample images. The OCR software assigns standardised codes to the recognised letters, which can then be used in data processing.

Through these processes, OCR enables texts to be further processed and analysed in various computer programs, which significantly improves efficiency and accuracy when managing and processing large volumes of documents.

Please note: Unfortunately, there are always misunderstandings regarding the terminology in discussions about the question "What is OCR?" between specialist departments or with our customers (see the distinction between OCR, iOCR and AI).

Table of contents

OCR ist die Basis für die Prozess-Automatisierung – dank BLU DELTA KI sogar in der Interpretation der Bedeutung

OCR ist eine Technologie, die die Umwandlung von gescannten Papierdokumenten, PDF-Dateien oder Digitalfotos in bearbeitbare Dokumente für Computer und Software (wie Microsoft Word oder FIBU-Software) ermöglicht. Damit lassen sich selbst Einzelposten extrahieren, wie in diesem Blogbeitrag "Einzelposten erfassen mit OCR" nachzulesen ist.

Die Geschichte der OCR reicht zurück bis in die 1920er-Jahre, als die ersten Ansätze zur maschinellen Texterkennung entwickelt wurden. In den folgenden Jahrzehnten haben sich die Technologien stetig weiterentwickelt, wobei in den 1970er-Jahren die ersten kommerziellen OCR-Systeme/Scanner auf den Markt kamen. Ein bedeutender Fortschritt war die Einführung von maschinellem Lernen und neuronalen Netzen in den 2000er-Jahren, die die Genauigkeit und Effizienz der Texterkennung erheblich verbesserten.

Wenn Sie ein Dokument in Papierform haben – zum Beispiel eine Rechnung, eine Bestellung oder einen Vertrag, den Ihnen jemand als PDF-Anhang geschickt hat – reicht ein Scanner allein nicht aus, um mit den relevanten Informationen aus diesen Dokumenten zu arbeiten. Der Scanner macht nur ein Bild des Dokuments, das aus einer Ansammlung von Bildpunkten besteht. Zur Weiterverarbeitung der Informationen aus gescannten Dokumenten, Digitalbildern oder Bild-PDFs benötigen Sie eine moderne OCR-Software/Anwendung zur Texterkennung. Denn die erkennt in dem jeweiligen Bild sämtliche Zeichen, setzt diese zu Wörtern und Zahlen zusammen und generiert daraus ganze Sätze. Damit erstellt die Software aus einem Bild eine Zeichenkette, einen Text.

Seitdem Deep Learning im OCR-Bereich angewandt wird, hat die Qualität der Texterkennung stark zugenommen und ist auf Augenhöhe mit der menschlichen Erkennungsfähigkeit. Durch den Einsatz von Deep Learning kann die OCR-Technologie nicht nur Zeichen und Wörter präziser erkennen, sondern auch komplexere Layouts und Schriftarten besser verarbeiten. Erfahren Sie hierzu im Detail gerne in unserem OCR vs. DeepOCR-Vergleich mehr.

Nun fehlt aber noch die semantische Bedeutung des Texts und der Zahlen (z. B. "Welche Zahl ist der Bruttogesamtbetrag?"), damit Sie Ihre Prozesse ohne "human in the loop" automatisieren können. Und genau hier kommen wir in Spiel: Wir setzen auf fortschrittliche Algorithmen und künstliche Intelligenz zur Texterkennung, wodurch der Kontext und die Bedeutung der erkannten Zeichen automatisch interpretiert werden. Dies ermöglicht eine vollautomatische Verarbeitung und Analyse der Dokumente, was die Effizienz und Genauigkeit in der Datenverarbeitung erheblich steigert. So lässt sich OCR z. B. optimal für Rechnungen und viele andere Dokumente (siehe auch OCR-Belegerfassung) nutzen.

How does our OCR text recognition software work? What is the advantage of combining OCR & AI?

To understand how OCR software works to recognise all characters, let's take a look at the various steps involved in text recognition. As already mentioned at the beginning of this text, the OCR application first analyses the structure of the document. To do this, the technology divides the page into text blocks, tables and images. These are then divided into lines, which in turn are broken down into words and finally into individual letters. Once the letters have been identified, the programme compares them with a series of sample images and calculates the probability of a match (for example, a character could be recognised as "A" 89% of the time). The OCR software then decides in favour of the character with the highest match.

A modern OCR system such as our software can also be configured for multiple languages. In addition, many OCR systems, including our artificial intelligence for text recognition, offer dictionary support for different languages. This support can be particularly useful when optimising OCR for specific domains, such as accounting. The integration of specialised dictionaries and specific terms can significantly improve the accuracy of text recognition in a particular context.

A major advance in OCR text recognition is the integration of artificial intelligence (AI), deep learning and large language models (LLM). This is because AI-supported systems use neural networks trained by deep learning to recognise patterns and fonts with greater precision. These systems for LLM data capture are able to reliably process even complex layouts and varying fonts and offer significantly higher recognition accuracy than traditional OCR technologies.

Another important aspect is the difference between pre-trained OCR systems and those that need to be individually trained. Pre-trained OCR systems are ready to use and offer excellent performance for general applications. They are optimised for a wide range of fonts and layouts and can be implemented quickly. Individually trained systems, on the other hand, require specific customisation to a company's needs, which requires additional time and resources for training and adaptation.

Overall, it is clear that the further development of OCR technologies through the use of AI, deep learning and LLMs has significantly expanded and improved the possibilities of text recognition and document capture. And this is precisely why we rely on these new technologies to provide you with optimum support in data extraction!

Image quality is crucial for automation with OCR

Text recognition from an image and the associated conversion into a document only takes a few seconds. As a result, the first step is to obtain a text and its meta information relating to text size, font and position without any manual effort.

This information now makes an image searchable and editable. However, the semantic meaning of the text is of course still required for comprehensive automation. OCR and automated text recognition are therefore important cornerstones for the automation of your processes - but not everything! This is because the characters, words and numbers and their meta-information form an important data source for algorithms and AI models based on them, which assign semantics to the jumble of letters.

Our BLU DELTA KI invoice capture system uses the results of the OCR to automatically extract valuable information for subsequent processes (e.g. accounts payable) without any further manual effort. You not only receive character strings, words and numbers, but also their meaning.

As already mentioned, the OCR software determines the probability of how closely a character corresponds to a specific number or letter. This probability varies with the image quality. Blurred images, text with a coloured background or simply poorly scanned documents can have a major impact on quality. In our regular BLU DELTA benchmarks (quality measurement at KI), we see that the photo and scan quality is decisive for the subsequent processes.

An "8" quickly becomes a "6" or a "B". However, a "tilted" letter has no effect on our automation. Modern NLP (Natural Language Processing) approaches, such as those we use at BLU DELTA, reduce such individual errors.

Up to 30 % higher automation rate

Due to poor scan and image quality, we see differences of up to 30 % in our customers' automation rates in document capture. A distinction is made between digital photo, scan and PDF text in terms of input quality. These differences are also a reason why we at BLU DELTA offer a prediction of the automation rate for invoice capture.

Digital photo and OCR

As a rule, images taken with mobile devices have the following problems:

• Shadows
• Uneven illumination
• Incorrect perspective
• Additional areas outside the page borders

OCR software can correct these problems to a certain extent. Nevertheless, digital photos pose the greatest challenge for automation due to the points mentioned above. So-called CamScanners or similar mobile OCR scanners and/or image optimisations can improve the quality accordingly in advance.

Scan and OCR

Professional scanners already provide a good basis for the automated processing and capture of documents. If possible, scan your documents in black and white (so that loss-free compression is possible) and with at least 300 dpi. Small fonts up to 9pt can still be easily recognised.

PDF text and OCR

PDF text delivers the best results. The actual OCR process is usually omitted here. The PDF document already contains the characters in digital form and the subsequent process "only" has to recognise the semantics. Documents in pure PDF text format achieve overall recognition rates of more than 90 % with BLU DELTA AI. If possible, you should therefore ensure that you receive unstructured or semi-structured documents as PDF text from your document sources.

However, PDF text documents are also often enriched with images containing text information. This relativises the advantage in this case.

Different types and application areas of OCR

Optical character recognition is a versatile technology that can be used in various forms and for a wide range of applications. There are two main types of OCR systems: Text recognition and handwriting recognition (ICR). Text recognition is used to extract printed text from digital images, scans or PDFs, while handwriting recognition aims to convert handwritten notes or documents into machine-readable text.

Particularly in the field of (accounts payable) accounting, the term OCR is often equated with the capture of information from invoices. From a technical point of view, however, this is a separate process. BLU DELTA AI contains a component for text recognition and, based on this, AI models that capture the semantic relationships.

OCR is used in numerous industries:

• In accounting, OCR is used to digitise and process invoices and receipts.
• In healthcare, OCR enables the fast and accurate capture of patient data and medical records.
• In logistics, OCR helps with the management and tracking of delivery documents and shipment tracking.
• Insurance companies use OCR to automate claims processing.
• In finance and banking, OCR enables the efficient processing of transactions and documents.
• OCR is also used in the real estate sector to digitise documents such as rental agreements and property deeds.

BLU DELTA AI can be used for text recognition via cloud or on-premise

The choice between on-premise and cloud-based OCR solutions often depends on the specific requirements of the industry and data security needs. Both are possible with our software. If you opt for the on-premise version, this is installed locally on your company's servers and offers a high level of control over data and processes, but is associated with slightly higher initial costs and more maintenance work. If you opt for the cloud solution, this enables flexible and scalable use.

On the subject of data security, in the context of information security management systems (ISMS) and the General Data Protection Regulation (GDPR), OCR systems must be configured in such a way that they comply with the applicable data protection and security requirements in order to guarantee the confidentiality and integrity of the processed data. It goes without saying that both our versions fulfil this requirement.

Conclusion - OCR: Paving the way for efficient document processing

Optical Character Recognition (OCR) is a powerful technology for converting scanned documents, images and PDFs into machine-readable text data. By analysing and interpreting text structures, OCR combined with artificial intelligence enables efficient automation and processing of information in various industries such as accounting, healthcare, logistics, insurance and finance. The continuous development of technologies such as deep learning has significantly improved the accuracy and flexibility of OCR systems by reliably recognising both printed and handwritten text. While on-premise and cloud-based OCR solutions offer different benefits and requirements, the choice of the appropriate solution depends on the specific needs and security requirements of each industry. Overall, OCR is an essential foundation for digital transformation and increased efficiency in document processing.