| |
Methods
87
and cv are saved and displacement values from a range are tested for the next control point
until all relevant control points have been dealt with. At that time, the next iteration can start.
5.2.6.1 Rotating Wedge Visual Stimulus –“Grad. (Angular Dir.)”
The energy can be minimized in many different ways. The algorithm that was implemented
makes use of the structure of the data. The algorithm chooses one control point at a time and
assigns test displacement values to that control point. For every test displacement value, the
energy is calculated. If the energy has decreased, the displacement value is stored.
The order in which the control points are tested influences the result of the process. The
control points can be ordered in different ways. For example, a systematic order, starting with
linear index one, and finishing with the last linear index is possible. The control points can be
arranged randomly. In this case, the order should be saved so that the result can be repro-
duced. This leads to different results every time the program runs. Spirals, i.e., starting in the
middle of the image and ending in the periphery were also considered. It was decided to test
control points in regions with a high signal-to-noise ratio (high coherence) first so that noisy
regions do not lead the minimization algorithm into local minima at the beginning, from
where it might be difficult to escape later when the more reliable control points are tested.
Therefore, the order of control points is determined based on a 2D median filtered coherence
matrix. Coherence is seen as a confidence measure. Areas in the phase images that are very
noisy have lower coherence values than areas that are less noisy. Since noisy regions – which
can especially be found outside of V1, V2, and V3 – are not as important as the regions V1,
V2, and V3, control points outside of V1, V2, and V3 are tested once the others have been
already tested.
After the order of control points has been determined, derivatives in horizontal and vertical
directions of the deformed wedge-atlas image (at the beginning the initially aligned atlas) are
calculated. The resulting images are processed by using a maximum filter to deal with discon-
tinuities at the boundaries between adjacent visual areas. The gradient and its magnitude at the
|  |
|
| |
|
|
