To obtain a sample of each region large enough to statistically model its behaviour, initial seeds have to be placed completely inside the regions. Boundary information allows us to extract these positions in the ``core'' of regions by looking for places far away from contours.
Boundaries between colour texture regions, which are combination of colour edges and texture edges can be considered as perceptual edges, because a human has the ability to detect both ones. The problem of texture edge detection is considered as a classical edge detection scheme in the multidimensional set of texture features which are used to represent the region characteristics . Meanwhile, the extraction of colour boundaries implies a major difficulty since the use of an edge detector over a colour image produces the apparition of microedges inside a textured region. Our approach, is based on the perception of textures as homogeneous colour regions when they are seen from a long distance . A smoothing process is progressively performed starting from the original image until textures looks homogeneous, as we would look the texture from far away. Then, the application of an edge detector allows to obtain the colour edges. Figure 2 shows the effect of smoothing a textured image; regions which were originally textured are appreciated as homogeneous colour regions.
The union of texture and colour edges provides the perceptual edges of the image. Nevertheless, due to the inherently non-local property of texture and the smoothing process performed, the result of this method are inaccurate and thick contours (see Figure 2.c). However, this information is enough to perform the seed placement in the ``core'' of regions, which allows to model the characteristics of regions.