UNSUPERVISED SEGMENTATION OF SURFACE DEFECTS WITH SIMPLE TEXTURE MEASURES
Jukka Iivarinen
In M. Pietikäinen and H. Kauppinen (Eds.),
Workshop on Texture Analysis in Machine Vision, pages 53-58,
Oulu, Finland, June 14-15 1999.
Abstract
In this paper a simple and fast approach to unsupervised segmentation
of surface defects is described.
A set of simple texture measures (the local binary pattern method)
and the statistical self-organizing map are used to detect defects in
surface images.
Defect detection can be seen as a typical two-class segmentation
problem where the desired classes are good surface and defected surface.
The main idea in this paper is to train a two-class classifier only with
fault-free surface samples which is an unexpected approach.
The proposed defect detection is based on the following idea: an unknown
sample is classified as a defect if it differs enough from the estimated
distribution of fault-free samples. The statistical self-organizing map
is used to estimate this distribution. Surface images are used to
demonstrate the proposed procedure.
|