Image Processing Reference
FIGURE 5 Curve evolution under the proposed shape prior only.
This simulation shows that the proposed shape priors can well constrain an active contour
to take a given shape (known as reference) and handling nontrivial geometric shapes with
holes and complex topologies.
4.2 Application to Object Detection
We will devote this section to present some satisfactory results obtained by the proposed mod-
el in the case of Euclidean transformations, then the general case of the affine ones will be
treated. The algorithm is as follows: We first evolve the active contour without shape prior
until convergence (i.e., w = 1) to reduce the computational complexity and to have a good es-
timation of the parameters of the geometric transformation as in Refs. [ 8 , 11 ]. This first result
provides an initialization for the model with prior knowledge. Then the model will evolve un-
der both forces (data and prior forces) with more weight assigned to prior knowledge (gener-
ally w ≤ 0.5) to promote convergence toward the target shape.
4.2.1 Case of Euclidean transformation
To assess the performance of the method, we have experimented the proposed method in
medical imaging. The template is given by image 1 of Figure 6 which corresponds to the left
ventricle of the heart. The aim is to detect the true contours of the left ventricle of the heart
which are partially occluded by the valve. The second row of Figure 6 shows several iterations
of curve evolution until convergence without prior knowledge. Starting with final this result,
and based on the reference shape after rigid motion estimation, the evolving front continues
its evolution until the curve reaches the desired contours (third row).
FIGURE 6 First row: The template, test image, initial curve. Second row: The active contours
without shape prior. Third row: The active contours with shape prior.
The aligned curves which are used to compute the prior energy are presented in Figure 7 .
We present by Table 3 , the estimated rigid motion parameters.