Preliminary communication
Adaptive Image Processing Technique for Quality Control in Ceramic Tile Production
Snježana RIMAC-DRLJE
; Faculty of Electrical Engineering, J.J. Strossmayer University of Osijek, Osijek, Hrvatska
Drago ŽAGAR
; Faculty of Electrical Engineering, J.J. Strossmayer University of Osijek, Osijek, Hrvatska
Slavko RUPČIĆ
; Faculty of Electrical Engineering, J.J. Strossmayer University of Osijek, Osijek, Hrvatska
Abstract
Automation of the visual inspection for quality control in production of
materials with textures (tiles, textile, leather, etc.) is not widely implemented.
A sophisticated system for image acquisition, as well as a fast and efficient
procedure for texture analysis is needed for this purpose. In this paper the
Surface Failure Detection (SFD) algorithm for quality control in ceramic
tiles production is presented. It is based on Discrete Wavelet Transform
(DWT) and Probabilistic Neural Networks (PNN) with radial basis. DWT
provides a multi-resolution analysis, which mimics behavior of a human
visual system and it extracts from the tile image the features important
for failure detection. Neural networks are used for classification of the
tiles with respect to presence of defects. Classification efficiency mainly
depends on the proper choice of the training vectors for neural networks.
For neural networks preparation we propose an automated adaptive
technique based on statistics of the tiles defects textures. This technique
enables fast adaptation of the SFD algorithm to different textures, which
is important for automated visual inspection in the production of a new
tile type.
Keywords
Automated visual inspection; Discrete wavelet transform; Probabilistic neural network; Quality control in tile production
Hrčak ID:
56748
URI
Publication date:
30.4.2010.
Visits: 2.461 *