Multiscale Segmentation Techniques for Textile Images

In this chapter we focus on image segmentation techniques for some very special images — textile images. They are generated from the color halftoning technique in textile and printing production lines. In contrast with natural color images, textile images have some very distinctive features: (1) generally there are a few dominant colors in a textile image, whereas there may exist hundreds of significant colors in a natural color image; (2) in the textile industry designers combine threads of different colors, thickness, and densities to produce the visual impression of other colors (the color halftoning technique in textile printing). The fabric texture structure — texture noise — has a great influence on the colors’ appearance in textile images, which makes the color segmentation of textile images a very difficult problem [1]. A typical textile image is shown in Fig.1. According to human visual observation, there are about six dominant colors in the textile image, but when the image is enlarged to the pixellevel, we will find that there are many different colors in perceived uniform color regions as shown in Fig.1. (a) and (b). Four histograms of the spatial variation magnitudes of the four uniform and edge blocks (as shown in Fig.1 (a), (b), (c) and (d)) are illustrated in Fig.2 (Here the spatial variation magnitude is computed as the added value of the two absolute forward differences along the horizontal and vertical directions on the luminance component of each