Attribute combinations for image segmentation

Attribute multiplication

One approach, suggested by Lomask (2007), is to combine multiple attribute volumes into a single volume via multiplication:

$$A = \prod_{i=\mathrm{each attribute}}^{\mathrm{all attributes}} a_i ,$$ (1)

where $a_i$ is an individual attribute volume, and then proceed with segmentation normally. Multiplication of the attribute data has the effect of reinforcing information in areas where the attributes ``agree," which can be beneficial. However, it also can have the effect of destroying potentially valuable information if the two attributes are not in agreement. Panel (a) in Figure 4 shows the boundary calculation resulting from this process.

Clearly, in this case the disadvantages of multiplying attribute volumes together outweigh the possible advantages - the process appears to have incorporated the worst information from each of the attributes, resulting in a final boundary that does not improve on either of the individual results (Figure 3) in any location.

Attribute combinations for image segmentation