The data to be classified consists of a set of Brodatz textures. The texture images were bisected into the lower and the upper part. The upper part was used for training and the lower part for testing. The training and testing parts were further divided into a set of overlapping patches of fixed size (). Each patch was described by a feature vector its entries contain values of the co-occurrence matrices computed from the patch. The feature vectors computed on the patches of the upper part of the image form the training set. Similarly, the features computed on the lower part patches form the testing set.
Matlab scripts used to create the classification data:
brodatz_make | Computes feature description of Brodatz textures (*.tiff). |
brodatz_data1 | Make classification data from textures 1 and 2. |
brodatz_data2 | Make classification data from textures 5, 9 and 13. |
brodatz_data3 | Make classification data for Anderson's task. |