Articles
UNSUPERVISED IMAGE SEGMENTATION WITH NEURAL NETWORKS
Article number
562_10
Pages
101 – 108
Language
English
Abstract
The segmentation of colour images (RGB), distinguishing clusters of image points, representing for example background, leaves and flowers, is performed in a multi-dimensional environment.
Considering a two dimensional environment, clusters can be divided by lines.
In a three dimensional environment by planes and in an n-dimensional environment by n-1 dimensional structures.
Starting with a complete data set the first neural network, represents an n-1 dimensional structure to divide the data set into two subsets.
Each subset is once more divided by an additional neural network: recursive partitioning.
This results in a tree structure with a neural network in each branching point.
Partitioning stops as soon as a partitioning criterium cannot be fulfilled.
After the unsupervised training the neural system can be used for the segmentation of images.
Considering a two dimensional environment, clusters can be divided by lines.
In a three dimensional environment by planes and in an n-dimensional environment by n-1 dimensional structures.
Starting with a complete data set the first neural network, represents an n-1 dimensional structure to divide the data set into two subsets.
Each subset is once more divided by an additional neural network: recursive partitioning.
This results in a tree structure with a neural network in each branching point.
Partitioning stops as soon as a partitioning criterium cannot be fulfilled.
After the unsupervised training the neural system can be used for the segmentation of images.
Authors
J. Meuleman, C. van Kaam
Keywords
Neural networks, unsupervised learning, recursive partitioning, image segmentation
Online Articles (50)