Biomedical Engineering Reference
In-Depth Information
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k)
(l)
(m)
(n)
(o)
(p)
(q)
(r)
(s)
(t)
(u)
(v)
(w)
(x)
Figure 25.
Segmentation using shape prior active contour: (a-f) images for training; (g-l)
aligned training images; (m-r) eigenvectors after principal components analysis; (s) overlap
before aligning; (t) overlap after aligning; (u) zero level set of the mean shape; (v) level
set of mean shape; (w) target image to be segmented with initial contour; (x) segmentation
result. See attached CD for color version.
some heuristics. Ordinary gradient descent computes the direction of the steepest
descent by implicitly assuming a Euclidean metric on the weight space.
The Rprop algorithms are among the best performing first-order batch learning
methods. There are many advantages for Rprop methods:
1. Speed and accuracy.
2. Robustness.
3. First-order methods; therefore, time and space complexity scales linearly
with the number of parameters to be optimized.
4. Only dependent on the sign of the partial derivatives of the objective func-
tion and not on their amount; therefore, they are suitable for applications
