[Univ of Cambridge] [Dept of Engineering]
INDEX RESEARCH PUBLICATIONS DATABASE RESUME

WonUK_GTDB, a Remote Sensing Groundtruth Database for multiclass classification


WonUK_GTDB is an aerial image database we make available for research purposes. It is designed for testing image classification algorithms. The provided groundtruth information (semantic labels associated to images) can be used for evaluating the algorithm performance.


1. Database description

It consists of 1040 64x64 pixel images which were manually assigned to one of the following 8 classes:
  • boat
  • building
  • field
  • grass
  • river
  • road
  • tree
  • vehicle
These classes correspond to both manmade (building, boat, vehicle, road) and natural (field, grass, river, tree) entities. It is split into a training and a test set of 520 images each. Each set contains 8 subsets of 65 labeled images. Class sizes are summarised in the table below:

classbuildingboatfieldgrassriverroadtreevehicleall classes
training set 65 65 65 65 65 65 65 65 520
test set 65 65 65 65 65 65 65 65 520
total 130 130 130 130 130 130 130 130 1040


Click the image to see a sample of the database content:


2. Groundtruth information

For each image, its class label is given the name of the folder containing the image. For instance, the image test/tree/lmw_1878.bmp is a test image labeled as "tree". Alternatively, the groundtruth information is also stored in files "train_groundtruth_labels.txt" and "test_groundtruth_labels.txt" in the following plain text format:
     image_name    class_label

For example, in the file "test_groundtruth_labels.txt", we have the following entry:
     lmw_1878.bmp    tree

3. Image format

Images are 64x64 pixel, 24-bit color bitmap (.bmp) format.


4. Source

The 6 original aerial images were distributed by the British National Space Centre (BNSC) as a CDROM called "Window On The UK" in the Sunday Times on April 9th 2000. They cover both urban and rural areas of the United Kingdom.

For information a reduced version of the original images can be seen:



5. Credits

If you plan to use this database, please cite the following paper:
  • Julien Fauqueur, Nick Kingsbury and Ryan Anderson, Semantic discriminant mapping for classification and browsing of remote sensing textures and objects, accepted in IEEE International Conference on Image Processing (ICIP'2005). Genova, Italy, September 2005
Contact person: Julien Fauqueur ().

This work has been carried out with the support of the UK Data and Information Fusion Defence Technology Centre.


6. Download


J. Fauqueur - last updated: 7 june 2005