Recognition of clinker phases by automatic image analysis

Abstract

Microscopical analysis of clinker phases provides useful informations about its structure and its manufacturing process. This paper deals with automatic segmentation of clinker phases images acquired on polished sections using optical microscopy. The main difficulty lies in the segmentation step. Even sophisticated techniques using textural characterization did not give satisfactory results. Indeed, the concerned textures show variations from an image to another and their spatial extent may vary as well. The most efficient methods we have developed combine automated thresholding algorithms and the use of contrast. This latter notion is defined within the scope of a mathematical model: the logarithmic image processing (LIP) model. Considering the limited reproductibility of the chemical etches used to bring the phases to the fore and the variations in terms of shape and hue of the elements to be segmented, the proposed algorithms have to be adapted to clinkers of various origin. However their efficiency has been proved for one type of clinker.