Segmentation of rust defects on painted steel surfaces by intelligent image analysis

Abstract

Abstract Automated assessment of degree of rusting of the painted surfaces of metal constructions is an important task. It is complicated by the fact that the paints have different colours. Also, poor lighting due to its inhomogeneity and presence of shadows limit credible assessment of corrosion damage of steel surfaces. We propose an intelligent image analysis approach that solves above mentioned problems. Image processing is fulfilled in HSV colour space. In the case when the coating is red, a statistical segmentation is used. A parameterized Retinex method as image pre-processing was developed for coatings of other colours. Then alpha-matting is applied to provide adaptive segmentation of colour images and to detect rust stains on the image. These methods increase reliability of corroded area segmentation. The obtained results of segmentation are compared with the expert assessment.