Precise and Flexible – Powerful Text Recognition with Deep Learning Technology
The “Deep OCR” module in uniVision 3 combines high detection performance with immediate operational readiness – with no complex parameterization. Different fonts and sizes, irregular spacing, different orientations and damaged characters are no obstacle to text recognition. Compared to traditional OCR technology, Deep OCR also recognizes text, regardless of whether the text appears on a light or dark background. The strong performance is based on the proven Deep OCR technology of MVTec HALCON and ensures fast and reliable results even in complex requirements.
Ready-to-Use Technology
Thanks to a fully trained neural network, it is not necessary to create your own training data or a separate network training – the module is ready for use immediately with just a few settings.
Flexible Adjustment
Deep OCR reliably recognizes a wide range of fonts, languages and special characters. Long words and number sequences can also be reliably recognized and read by parameter adjustments in the module.
Robust Text Recognition
Even in poor lighting conditions, blurred, distorted or damaged text and on changing backgrounds, text recognition remains stable and precise.
High Speed
High-performance execution through the use of NPUs (Neural Processing Units) and dedicated inference frameworks to accelerate deep learning models. These are used in both the B60 smart camera and the machine vision controllers of the MVC series and enable inference times in the range of few milliseconds – depending on the number and size of the regions.
Integrated Heatmap Function
The heatmap is a visual representation that highlights areas of the image that the model has identified as letters or numbers. This makes the deep learning technology transparent and allows you to visually understand which features led to the decision. In addition, the module delivers a score value for each recognized word, which quantitatively evaluates the quality of text recognition and thus ensures maximum transparency.
Deep OCR Optimally Integrated in uniVision 3
The uniVision 3 image processing software is expanded by the “Deep OCR” module with powerful deep learning functions. By seamlessly integrating rule-based and AI-powered image processing, you benefit from a powerful, easy-to-use solution with optimal hardware support.
The combination of Deep OCR with the proven modules of uniVision 3 opens up completely new application possibilities. With the help of the “Image Region” module, image areas can be specifically excluded from text recognition, which significantly shortens the process time. At the same time, tracking modules such as “Image Locator” or “Image Pattern Match” allow these regions to remain automatically aligned with the relevant text for moving or variably positioned objects. With the help of the modules for 1D and 2D code reading, it is also possible to capture one or more codes within a single image in addition to texts. For further processing, the detected texts in the “Spreadsheet” module can be specifically analyzed using regular expressions (RegEx) and reduced to relevant information.
The combination of Deep OCR with the proven modules of uniVision 3 opens up completely new application possibilities. With the help of the “Image Region” module, image areas can be specifically excluded from text recognition, which significantly shortens the process time. At the same time, tracking modules such as “Image Locator” or “Image Pattern Match” allow these regions to remain automatically aligned with the relevant text for moving or variably positioned objects. With the help of the modules for 1D and 2D code reading, it is also possible to capture one or more codes within a single image in addition to texts. For further processing, the detected texts in the “Spreadsheet” module can be specifically analyzed using regular expressions (RegEx) and reduced to relevant information.
Hardware Advantages in Detail
B60 Smart Camera
- The integrated neural processing unit (NPU) enables Deep OCR to be executed on the B60 within milliseconds.
- Even demanding OCR tasks are processed quickly and accurately directly on the smart camera.
MVC Machine Vision Controller
- Supported by deep learning acceleration based on Intel’s OpenVINO™ technology.
- The computing load is intelligently distributed to the CPU and integrated GPU.
- This achieves maximum efficiency with minimal consumption of resources.
- Support for multiple cameras per controller is ideal for test stations with multiple interrogations at one position.
Deep OCR Applications
In many industrial applications, such as logistics, food, packaging and pharmaceutical industries, characters on different surfaces and materials must be accurately detected. The “Deep OCR” module guarantees precise recognition of dynamic text with variable font sizes and different backgrounds, even with specifically designed prints and labels.
Best Before Date/Label Recognition
The “Deep OCR” module reliably detects shelf life data and other label information. Poor contrasts, variable font sizes and obliquely applied labels do not affect text recognition.
Identification of Parts and Serial Numbers
In production processes, serial numbers can also be read accurately on metallic objects. The readability remains very high even in the event of contamination or reflection.
Recognition of Batch Numbers
The “Deep OCR” module can be used to read batch numbers on different packages. The detection rate is high even in the event of scratches or pressure damage.
Compatible Hardware
Deep OCR in wenglor uniVision 3 can be implemented with the following wenglor image processing products.
