Biomedical Engineering Reference
In-Depth Information
Fig. 4.6
Example of the Hough Transform for lines. Original image (
left
) and voting space (
right
)
example, Fripp et al. [
109
] proposed a cartilage initialization system that exploits
prior knowledge of the bone location.
4.8.3 General Hough Transform
In 1962, Hough [
110
] was granted a patent for
Method and Means for Recognizing
Complex Patterns
. His approach uses templates and a voting space. For each iden-
tified image feature all templates create votes for possible poses. The highest vote
represents a set of parameters of the best corresponding template. An example with
original image and voting space is shown in Fig.
4.6
. Illingworth and Kittler [
111
]
have presented a review of the Hough transform and Khoshelham [
112
] has demon-
strated an extension to 3D object detection.
The general Hough transform can be used for the initialization of deformable
models. Van der Glas et al. [
113
] have demonstrated a method to detect ball joints.
Seim et al. have proposed approaches for the hip [
114
] and the knee [
115
]. In
2010, Ruppertshofen et al. [
116
] have proposed a discriminative approach for lower
extremities, which has been extended to a multi-level scheme [
117
].
4.8.4 Atlas Registration
Rather than using a reference model, a reference image (called atlas) can be used. The
atlas contains a specific initialization. A registration step tries to find the transform
T
,
that matches the reference image to the actual image.
T
can then be used to transform
the initialization to the actual image.
Atlas-based initialization requires a registration process, e.g. using ElastiX [
118
].
Another requirement is the subjects pose, which has to be same in each image to
get satisfactory results. Examples for atlas registration based initialization have been
proposed by Fripp et al. [
119
,
120
].
