Biomedical Engineering Reference
In-Depth Information
Fig. 10.1 Samples of metal
surface after: a) sandpaper
polishing, b) turning, c)
milling, d) polishing with file
Feature
Extractor
Encoder
Classifier
Fig. 10.2 Structure of RSC
Neural Classifier
Fig. 10.3 Images of metal surface after milling
Fig. 10.4 Images of metal surface after polishing with sandpaper
own test set of metal surface images. We work with three texture classes, each
containing 20 images. From these 20 images, we randomly select some for training
the RSC, and the rest are used for testing. The number of images in the training set
ranges from three to ten.
The first texture class corresponds to the metal surface after milling (Fig.
10.3
),
the second to the metal surface after polishing with sandpaper (Fig.
10.4
), and the
third to the metal surface after turning with a lathe (Fig.
10.5
). You can see that
different lighting conditions greatly affect the grayscale properties of an image. The
texture may also be arbitrarily oriented, making the texture recognition task more
complicated.
