Detection of faults on rotary screen printed fabrics using machine vision
A project was sponsored by the SERC for research into the design of a colour vision system
for the detection of print faults in rotary screen printed fabrics. The research was carried out
at De Montfort University (formerly named Leicester Polytechnic), which has previous
experience with Image Processing in relation to Textiles. The proposed system was required
to identify, process and correct the common print faults which can occur during rotary screen
printing. These can be divided into two main categories, systematic and random faults.
This thesis covers the work undertakeni n the developmento f a laboratory-basedin spection
systema ndt he subsequendte velopmenat nd testingo f methodologiesto facilitate factory-based
on-line inspection. Initial investigation identified the requirement for colour segmentation
algorithmsa ndt he researchin to anda nalysiso f suitablem ethodologiesf or segmentationf orms
a fundamental part of this thesis.
Important, new colour segmentation algorithms were developed from first principles by the
author. These new methods offer improvements (in most cases significant) over the current
`state-of-the-art' colour segmentation technology, and are applicable to a wide-range of
computer vision tasks.
These proposed methodologies have been rigorously tested and the findings of the investigation
are presented as part of this thesis.