Page 87 - IJOCTA-15-3

Computational aspects of the crop field segmentation problem based on anisotropic active contour model

Figure 2. Variant of the contour lines in the cell

Figure 3. On the left: The original satellite photo. On the right: The photo after denoising

Figure 4. On the left: Histogram of the original image. On the right: Histogram of the
smoothed data

the following setting for given δ 1 , δ 2 > 0.
m = 3, l 0 = −20000, l 1 = −18000,
l 2 = −2000, l 3 = 10000, l 4 = 20000, However, it should be emphasized that the
σ = 3, τ = 0.1, η = 1 − ε, main drawback of the direct approach to the seg-
mentation procedure is the fact that the final re-
ε = 0.01, ∆t = 0.05,
sult (the shape of the obtained segments, their
and φ 0 ∈ C(Ω) of (41), the initial level set func-
area, and the corresponding coefficient of varia-
tion, was taken with S circle having radius 20 and tion) crucially depends on the number of segments
center in the central point of Ω, and d(x, S) the m+1 that we prescribe as an input characteristic.
Euclidean distance between the point x ∈ Ω and To circumvent this restriction, we make use of the
S. adaptive approach. In this context, we do not fix
The parameter α is defined by the rule (56). a priori the number of segments, but instead, we
Results of segmentation that have been obtained decompose the interval [l 0 , l m+1 ] into a uniform
in this way, are depicted in Fig. 5–6. In fact, grid
the given f-decomposition can be drilled down
l 0 < l 1 < l 2 < · · · < l M−1 < l M < l m+1
for each channel by iterative setting Ω = Ω i ,
i = 1, 2, and so on. Regarding the procedure’s with l j = l 0 + j∆l, where the integer M > 0 sat-
stop-condition, it can be taken as follows: isfies the condition
|Ω|
∗
2
either L (Ω i ) < δ 1 or CV f < δ 2 ≈ S .
Ω i M + 1
459