AI Solutions in Detail
The DENK Technologies
We explain what lies behind our technologies, the DENKnets. What distinguishes instance segmentation from image segmentation? For which use cases is object detection best suited? If you need more background information, feel free to contact us!
With suitable image analysis technology to precisely analyse
These DENKnets form the foundation for numerous industrial, medical, and logistical applications.
From simple classification to detailed segmentation – each method has specific use cases.
Image Classification
Image classification assigns an entire image to one or more categories. A neural network analyzes the image globally to classify it into a category, such as 'dog,' 'cat,' or 'defective component.' Image classification focuses on the entire image without localizing individual objects. It serves as the foundation for many advanced technologies, such as object detection or segmentation.
Possible applications:
- Quality control: Classification of defective and non-defective products.
- Medical diagnosis: Detection of disease patterns in X-ray or MRI images.
- Agriculture: Detection of plant health or crop conditions.
Object Detection
Object detection is a key technology in image processing. It identifies and locates various objects within an image using bounding boxes or rotated bounding boxes. It can determine not only positions but also categories. Unlike image classification, which assigns an entire image to a category, object detection specifically analyzes individual objects and their positions.
Possible applications:
- Logistics: Detection of packages or objects on conveyor belts.
- Smart Cities: Traffic monitoring, e.g., detection of vehicles, pedestrians, or traffic lights.
- Retail: Inventory tracking of products on shelves.
Image Segmentation
Image segmentation goes beyond object localization and classifies each pixel in an image. This creates precise masks that clearly distinguish individual objects or regions. This technology provides the highest level of detail in image analysis. Unlike bounding boxes and keypoints, it assigns a class to each pixel, allowing for detailed analysis of complex scenes.
Possible applications:
- Urban planning: Segmenting road elements such as lanes, sidewalks, and green spaces.
- Medicine: Highlighting tumor tissue or anomalies in medical scans.
- Agriculture: Analysis of plant areas or detection of weeds.
Instance Segmentation
Instance segmentation combines the concepts of image segmentation and object detection. Each object is not only classified but also individually masked, allowing multiple instances of the same class to be displayed separately. Unlike image segmentation, instance segmentation identifies individual objects within the same class. This makes it ideal for scenes with many overlapping objects.
Possible applications:
- Logistics: Individual package detection on conveyor belts.
- Biology: Segmentation of individual cells in microscope images.
- Manufacturing: Identifying, counting, or measuring individual workpieces on assembly lines.
OCR
OCR technology enables the recognition and extraction of letters, numbers, and (special) characters. Image data is analyzed, characters are detected, and the text is output in a structured format. Compared to other technologies, OCR works specifically at the text level. It is essential for applications where signs or text analysis are the main focus.
Possible applications:
- Food industry: Expiration date recognition.
- Traffic control systems: License plate recognition.
- Various industries: Product labeling.
Barcode and QR Code Reading
Barcode reading technology captures barcodes or QR codes in images and interprets their contents. Image sections are analyzed, and the stored information is decoded. Unlike traditional OCR, barcode reading focuses exclusively on analyzing encoded symbols. This process is faster and more efficient than general text recognition approaches.
Possible applications:
- Logistics: Automatic detection and tracking of packages.
- Retail: Inventory management and point-of-sale systems.
- Industry: Identification of components and parts.
