312
Q. Chen et al.
11.3.2.2 Iterative GA Segmentation
Considering the intensity inhomogeneity and high noise level typically present in GA
projection images (shown in Fig. 11.20a and b), conventional thresholding techniques
[68, 69] would produce masks that are too coarse to be considered as an adequate
initialization for the subsequent CVLSF model, because they frequently exclude
large portions of GA regions that cause convergence into local minima during the
refinement step. An alternative iterative threshold method based on global image
information is proposed here to coarsely segment GA regions from the projection
image. This iterative threshold is computed considering a restricted region within
the image that gets updated during subsequent iterations, and is set to decrease and
converge to a certain value (see later experimental results and analysis section).
(a)
(b)
(c)
(d)
(e)
Restricted region
(NR
2
)
Fig. 11.20 a GA projection image with GA contour generated by manual segmentation. b Histogram of GA region and background and the threshold using OTSU method over the whole projection image. c Segmentation result obtained by OSTU method in the first iteration. d Histogram
of the whole image where the mean value of the foreground region resulting from the first iteration
and the values corresponding to the restricted region for the second iteration are indicated. e Final
result for the coarse GA segmentation
Q. Chen et al.
11.3.2.2 Iterative GA Segmentation
Considering the intensity inhomogeneity and high noise level typically present in GA
projection images (shown in Fig. 11.20a and b), conventional thresholding techniques
[68, 69] would produce masks that are too coarse to be considered as an adequate
initialization for the subsequent CVLSF model, because they frequently exclude
large portions of GA regions that cause convergence into local minima during the
refinement step. An alternative iterative threshold method based on global image
information is proposed here to coarsely segment GA regions from the projection
image. This iterative threshold is computed considering a restricted region within
the image that gets updated during subsequent iterations, and is set to decrease and
converge to a certain value (see later experimental results and analysis section).
(a)
(b)
(c)
(d)
(e)
Restricted region
(NR
2
)
Fig. 11.20 a GA projection image with GA contour generated by manual segmentation. b Histogram of GA region and background and the threshold using OTSU method over the whole projection image. c Segmentation result obtained by OSTU method in the first iteration. d Histogram
of the whole image where the mean value of the foreground region resulting from the first iteration
and the values corresponding to the restricted region for the second iteration are indicated. e Final
result for the coarse GA segmentation