Related Experiment Videos
Automatic segmentation and classification of outdoor images using neural networks
N W Campbell1, B T Thomas, T Troscianko
1Advanced Computing Research Centre, University of Bristol, UK. Neill.Campbell@bristol.ac.uk
International Journal of Neural Systems
|February 1, 1997
Summary
This study demonstrates neural networks for image segmentation and object labeling. Using a self-organising feature map and multi-layer perception, 91.1% of image areas were accurately classified into eleven categories.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Image analysis often requires accurate object segmentation and classification.
- Traditional methods can be labor-intensive and may lack precision.
- Neural networks offer a promising approach for automated image understanding.
Purpose of the Study:
- To explore the application of neural networks for image segmentation and object labeling.
- To evaluate the effectiveness of self-organising feature maps for segmentation.
- To assess the contribution of color and texture features in classification.
Main Methods:
- Image segmentation was performed using a self-organising feature map.
- A multi-layer perception was trained for region labeling.
- The quality of segmentation and feature contributions were quantified.
Main Results:
- The combined approach achieved high accuracy in image segmentation.
- Color and texture features were shown to be significant contributors.
- The multi-layer perception correctly classified 91.1% of the image area into eleven categories.
Conclusions:
- Neural networks provide an effective solution for automated image segmentation and object labeling.
- The proposed method demonstrates robust performance across various object categories.
- This approach has significant potential for applications in image analysis and computer vision.