An affine shape constraint for geometric active contours - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 431

FIGURE 7 First image: Curves before alignment, Second image: Curves after alignment.

Table 3
The Estimated Motion's Parameters

Rotation (°) Scaling Difference Between Starting Points

− 0.802 1.571 1.312

We end this part of experiments by applying the active contour with shape prior to object
tracking. In Figure 8 , the goal is to track the mouse in a video sequence with clutered back-
ground. We suppose that the movement of the mouse through frames is rigid thus the presen-
ted method of shape alignment is used. The convergent contour in frame i is used as a tem-
plate in frame i + 1.

FIGURE 8 Tracking of the mouse in video sequence. First row: The active contours model
without shape prior. Second row: The active contours model with shape prior.

Results seem to be satisfactory. Although, the presented rigid motion estimation method
solves some situations and leads to interesting detection results. Other types of transforma-
tions such as stretching can only be solved by the class of relevant transformations. Thus the
goal of the next section is to present some obtained results within this framework.

4.2.2 Case of affine transformation
Consider for this first experiment, the spider object, image (a) of Figure 9 , obtained from MCD
database (url: htp://vision.ece.ucsb.edu/zuliani/Research/MCD/MCD.shtml ) . The (b) image is
the object of interest which is obtained from (a) after affine transformation and partial occlu-
sion. As a first step, we perform object (b) segmentation without prior knowledge. The ob-
tained contour is then aligned with the contour of the (a) object to determine the regions of oc-
clusions. We present in (c) the result of shape alignment. The estimated values are α = 2, l 0 = 0,
A = [0.5, 0.2, 0, 2]. By Figure 10 , we present the obtained results without and with shape prior.
We considered in this experiment four templates (a spider, a chopper, a device, and a bird),
see Figure 11 . We were based on the set of the presented invariant shape descriptors to choose
the suitable reference according to the occluded spider (image (b) of Figure 11 ) .

FIGURE 9 Spider's shapes alignment.