Classifying Textile Designs using Region Graphs

Markov random field pixel labelling is often used to obtain image segmentations in which each segment or region is labelled according to its attributes such as colour or texture [4]. This paper explores the use of such a representation for image classification. In particular, the problem of classifying textile images according to design type is addressed. Figure 1(a) shows an example of an image segmented into groups of regions by assigning each pixel a label; the label image is shown in the centre. Given such a labelling, the image can be represented as a bag of shapes by computing shape descriptors for each connected component [3]. However, a bag of shapes model ignores relationships between the groups of regions. In order to retain information about these relationships, we construct undirected weighted graphs as shown in Figure 1(b). Each vertex is associated with a group of regions (bag of shapes). Edges in the graph denote either the extent to which the groups’ regions are spatially adjacent or the dissimilarity of their respective bags of shapes.