Image Processing Reference
FIGURE 10 Detection without, image (a), and with, image (b), affine prior knowledge.
FIGURE 11 Image (a): The available templates. Image (b): The transformed object.
Table 4 presents the obtained Euclidean distance between the target shape and the available
templates. We notice that the minimum distance can be easily identified because the suitable
form in this experiment is distinguishable.
Distances Between the Occluded Spider and the Used Templates
Spider Chopper Bird Device
The occluded spider 0.037
In the next experiment, a mosaic application is considered. The mosaic images that are con-
sidered are taken from the Bardo Museum of Tunisia which contains the biggest collection of
mosaic images in the world. In mosaic images, objects are composed of tessellas and are often
FIGURE 12 Some mosaic images from the Bardo Museum of Tunisia.
So, given that in mosaic images the objects are often repeated and in order to study the
robustness of our method, we try to find to true contour of an occluded object based on an-
other one having the same shape. We have approximated the perspective projection to an af-
ine transformation which is often used in the literature according to the acquisition condi-
tions. Let's consider the (c) image. As it's shown, this image contains many forms. Among all
forms available in these two images, (c) and (d), taken from two sides, we use the Euclidean
distance between the affine invariant Fourier descriptors to localize two lions. The left one is
partially occluded. We will use the right lion of the (d) image as template in order to have a
better segmentation. In the last figure, images (a) and (b), we present the used curves to per-
form shape alignment and shape prior computation. The estimated values are α = 1, l 0 = − 0.52,
A = [− 0.002, − 0.003, 2.003, − 0.35]. We present by the (c) image the obtained curves alignment
result and by the (d) image the segmentation result based on the proposed shape prior. Such
results are particularly interesting since it can be used for mosaic images restoration under
partial occlusions and missing parts. The underlying idea is to extract similar forms using min-
imal distance between descriptors in order to define prior knowledge for such occluded or