OCR text recognition
Seamless error detection
and perfect readability
In order to automate the effort of textual error detection in production, reduce error rates and provide a reliable solution, elunic developed a system for automatic and seamless text recognition with the help of AI.SEE™.
Goals of automated detection
Solution
- Creation of an application that scans each object and the text on it
- Supply of unambiguous training data on the basis of which the rejection criteria are learned
- An adaptive learning model that is constantly evolving
- Correction of OCR results by analyzing the context (Intelligent Character Recognition)
- Line-by-line text recognition using neural networks (instead of classic character-by-character text recognition) for even shorter checking times
- Comprehensive data collection and analysis so that traceability is possible and measures can be taken (e.g. error-prone areas are found and eliminated)
- Real-time monitoring can be implemented in the existing system
- Accessible from any end device and data collection in the cloud
Background
Optical input devices, such as scanners, generate a pixel graphic when a page of text is scanned. This means that they do not recognize the displayed letters as such, but only dots arranged in columns and rows in different colors. The recognition of conventional characters - as opposed to fonts developed specifically for OCR purposes - is more complex and requires software that can quickly and without great computational effort distinguish the individual characters from one another, recognize them and reproduce them correctly or check that they are free of errors. Human verification cannot do this.
Text recognition is also known as optical character recognition (OCR). In IT, this refers to the automated recognition of text within images.
It is not only production that needs to be automated in order to remain competitive, but also quality assurance. Flawless texts are of great importance in industry (for example in blister inspection in the pharmaceutical industry), as errors not only make the product appear inferior but can also cause misinterpretation. Partial or complete illegibility can deprive the user of important information, resulting in safety gaps and/or legal problems. In addition, particularly fast and reliable error detection is crucial for efficient production.
AI.SEE™ uses artificial intelligence to enable manufacturers to carry out non-destructive, reliable and automated inspections of their products.
Thanks to deep learning algorithms, even the smallest defects on heterogeneous or reflective surfaces that are often overlooked by human inspectors can be found with consistently reliable quality, providing powerful insights that help improve quality and operational efficiency.
References
Find out more about the use case.
