Image Processing Reference
In-Depth Information
(a) Search space
First stage
contour
Final
contour
End
point
End
point
Start
point
Start
point
(b) First stage open contour
(c) Second stage open contour
Figure 6.9
Discrete dual contour point space
The technique was originally demonstrated to extract the face boundary, for feature
extraction within automatic face recognition, as illustrated in Figure 6.10 . The outer boundary
(Figure 6.10 (a)) was extracted using a convex hull which in turn initialised an inner and an
outer contour (Figure 6.10 (b)). The final extraction by the dual discrete contour is the
boundary of facial skin, Figure 6.10 (c). The number of points in the mesh naturally limits
the accuracy with which the final contour is extracted, but application could naturally be
followed by use of a continuous Kass snake to improve final resolution. In fact, it was
shown that human faces could be discriminated by the contour extracted by this technique,
though the study highlighted potential difficulty with facial organs and illumination. As
already mentioned, it was later deployed in cell analysis where the inner and the outer
contours were derived by analysis of the stained-cell image.
Snakes, or evolutionary approaches to shape extraction, remain an attractive and stimulating
area of research, so as ever it is well worth studying the literature to find new, accurate,
techniques with high performance and low computational cost. We shall now move to
determining symmetry which, though more a low-level operation, actually uses evidence
gathering in some implementations thus motivating its later inclusion.
6.4
Discrete symmetry operator
The discrete symmetry operator (Reisfeld, 1995) uses a totally different basis to find
 
Search WWH ::




Custom Search