Hysteresis
Thresholding, Edge Closure and Aggregation
In order to eliminate the edges which represent noise in the image,
each candidate edge is labeled with its gradient norm value and a hysteresis-based
thresholding method is applied. This algorithm gives well-connected edges
while eliminating isolated noisy edges. The threshold values are chosen
low enough to obtain a light over-segmentation.
Then, the resulting edges have to be closed to obtain clusters. The
opened chains are extended according to the direction of the chain extremity
and the value of the gradient module of the pixels with the use of an algorithm
based on A* algorithm. This rapid method is an extension of a previous
one described in [V. Bessettes, 1997]
and [B. Chebaro, 1993]. It will be
described later.
Finally, the small regions are grouped according to their average gray
level into the second band and the contrast of the boundaries, to obtain
significant samples for the classification process (minimal surface) (Figure
64). The following distance between two
regions is chosen:
With:
AvgG(R): average gray level of the region R,
CtrB(R1, R2):
average contrast of the boundary between the regions R1
andR2,
,
: positive
coefficients. 