Texture based duplex-board layer segmentation

Abstract: A serial scanning electron microscopy images of paper was obtained by a slicing procedure The slicing procedure leads to misregistrations that have to be corrected. Two methods for registration of consecutive images are presented here.Duplex-board is used mainly for packaging liquids. It has a triple layer structure. Separation of these layers, is of great importance not only for getting a better understanding of the structure of paper, but also for being able to measure and verify various optical and mechanical properties of it. The problem of separation of layers in paper is formulated as a texture classification task. Here we use statistical properties of textures for locating boundaries between the different layers in duplexboard. The density distribution of fibre points is used for locating borderlines which can separate the layers. We also used co-occurrence features in a classifier which discriminate the layers of the board. We investigate methods to reduce the dimensionality of the co-occurrence feature set. We introduce additional features, such as the density distribution of fibre points, and the distance of pixels to the surface of paper, and use them to achieve better classification results. The experiments show that co-occurrence based classification methods provide useful information about the complex structure of duplex-board.