Image Segmentation with Eigen-Subspace Projections - Video Segmentation and Its Applications - page 41

Fig. 2.13 SHVS search

algorithm

advantage of the SSS method is suitable to segment the interior part of the object
because its color is changing smoothly. But, the SSS method is not good at seg-
menting the exterior part of the object because the boundaries with different color
will be a major problem. The FQS method is searching separately in four quadrants
where each area is independent. The advantage of the FQS method is that different
color area will not influence each other. The disadvantage of the FQS method is that
small area in certain quadrant will be difficult to extract. The main reason is that the
eigenvectors are updating too fast and is hard to segment the abrupt color-changing
area. The SHVS method takes both advantages from the SSS and the FQS methods
that is most suitable to our simulation.

2.3.3 Block Diagram of Adaptive Eigen-Subspace

Segmentation Method

Figure 2.14 shows the block diagram of the proposed AESS method. First, we will
sample the chosen pixel and the surrounding eight pixels to obtain the initial eigen-
vectors as in Fig. 2.15 . Then, new eigenspaces, related to the pixel obtained by the
searching algorithms as in Figs. 2.11 , 2.12 ,and 2.13 will be formed. Then, we apply
the eigen-subspace transformation of the color planes to form the signal and noise
planes as shown in Fig. 2.16 . Finally, we can use ( 2.39 ) to differentiate between the
desired and unwanted color pixel. If the pixel is justified as a desired color pixel then
the AESS method will be applied otherwise the algorithm will search next pixel and
take the eigenvectors of current pixel as reference. We devise an equation to separate