Segmentation methods in ZEN
Machine Learning vs. Deep Learning
In the context of segmentation, classical machine learning and deep learning represent two distinct approaches to pixel classification and object identification.
- Classical Machine Learning: This approach relies on algorithms such as decision trees, random forests, or support vector machines to classify pixels based on manually selected features (e.g., intensity, texture, or shape). It requires feature engineering, where users define the relevant characteristics of the data. Tools like the Intellesis Trainable Class Segmenter in ZEN and ZEN core utilize classical machine learning to provide robust and customizable segmentation workflows.
- Deep Learning: In contrast, deep learning uses neural networks to automatically learn features from data during the training process. This method excels in handling complex datasets and identifying patterns that may not be apparent through manual feature selection. For instance, AI Instance Segmentation in ZEN and ZEN core applies deep learning models to distinguish and separate overlapping objects with high precision.
Both approaches are available in ZEN and ZEN core, allowing you to choose the method that best suits their data and analysis requirements. While classical machine learning is often faster to implement for simpler tasks, deep learning is ideal for more challenging segmentation problems involving complex or heterogeneous datasets.
|
Aspect |
Machine Learning (Intellesis) |
Deep Learning (AI Segmentation) |
|---|---|---|
|
Setup effort |
Low–Medium |
Medium–High |
|
Training location |
Inside ZEN |
External / cloud models |
|
Feature extraction |
Manual |
Automatic |
|
Transparency |
High |
Lower |
|
Performance on simple data |
Very good |
Good |
|
Performance on complex data |
Good |
Excellent |
Related Topics
See also
Choosing the right segmentation method
The following list shows you the recommended segmentation method for your given scenario:
|
Scenario |
Recommended Method |
|---|---|
|
High contrast, uniform intensity |
|
|
Uneven illumination |
|
|
Slightly touching objects |
|
|
Complex textures or signals |
Intellesis Segmentation / AI Semantic segmentation (arivis Cloud) |
|
Overlapping or dense objects |
AI Instance segmentation (arivis Cloud) |
Key Takeaways
- Segmentation converts images into measurable objects
- ZEN and ZEN core offer classical, machine learning, and AI-based segmentation methods
- Classical methods are fast and best for simple images
- Intellesis enables flexible, trainable segmentation directly in ZEN / ZEN core
- AI-based segmentation provides the highest accuracy for complex or overlapping objects
- Choosing the right method depends on image complexity and analysis goals