Segmentation of Partially Shadowed Rust Images

The use of digital image processing for the analysis of corroded painted surfaces of constructions is considered. Rust segmentation of painted surface images in different illumination conditions, especially shadows, is a challenging task. Shadows influence on RGB pixel values in images and lead to further oversegmentation. The parameterized Retinex is used to eliminate non-uniform illumination. The influence of color restoration of partially shadowed rust images on segmentation results is studied. The color restoration is fulfilled using ColorChecker and least-squares approximation by a linear function. Gaussian mixture models based rust segmentation was used. Finally the results of experiments are shown and analyzed.