An unsupervised application for the visual analysis of metallic substrates and paint coatings is presented. Model based classification is used for the analysis of metallic substrates while a segmentation of the image is required to distinguish between paint and substrate or between paint and rust for the analysis of paint coatings. A unified framework based on the Expectation-Maximization (EM) algorithm is presented to solve both analyses. The EM algorithm is used both as a colour-based identification method for measuring cleanness properties of the substrate's surface, and also as a colour-based segmentation method for extracting the paint regions of the coating's surface. An initialisation method and a pre-processing step are proposed to avoid false segmentations. Experimental results on real samples show more repeatability and accuracy than the results obtained by human visual inspection. Each analysis has been visually illustrated with a representative example.
|Title of host publication||IEEE International Conference on Image Processing|
|Number of pages||4|
|Publication status||Published - 14 Sep 2003|
|Event||Proceedings: 2003 International Conference on Image Processing, ICIP-2003 - Barcelona, Spain|
Duration: 14 Sep 2003 → 17 Sep 2003
|Conference||Proceedings: 2003 International Conference on Image Processing, ICIP-2003|
|Period||14/09/03 → 17/09/03|