Neural Computing Research Group BAe

Validation and Verification
of Neural Network Systems

The Database

British Aerospace provided a database (which is part of the Sowerby Image Database) consisting of 119 coloured images of outdoor scenes.

The RGB image is a 24-bit colour digitised picture that has been calibrated to eliminate various undesirable effects in the image acquisition stages such as the systematic error in the colour levels introduced during the digitisation phase.

The images have been ideally segmented. This process consists in the splitting of the image in several regions which roughly correspond to the objects (or parts of an object) contained in the scene. Then the over-segmented regions corresponding to one object have been merged, whereas under-segmented regions have been further split.

The regions have been hand labelled and classified in one of 11 categories.

n Class
1 Sky
2 Vegetation
3 Road Surface Marking
4Road Surface
5 Road Border
6 Building
7 Bounding object
8Road sign
9 Telegraph Pole
10 Illumination Shadow
11 Mobile Object

List of the 11 classes labelling the regions.

The following links display two of the images drawn from the data base:

Rural scene:

Urban scene:

We have described the whole database by computing a set of 35 features. A region has been characterised features describing its internal properties as well as the the context surrounding it.

Feature Internal characteristics
1 Colour
2-3 Hue
4-6 Topological features
7 Size
8-9 Position
10-16 Shape descriptors
17-19 Texture descriptor



Contextual characteristics
20-23 Relative intensities
24-27 Relative size
28-35 Relative position

List of the 35 features describing the regions of the database.

Project Home page Introduction Work plan Publications

Contact names

Francesco Vivarelli
Dr. Christopher K. I. Williams
Dr. W. Andrew Wright

This page is maintained by Francesco Vivarelli (vivarelf@aston.ac.uk)
Last modified: Thu Jun 26 19:56:20 BST