TITRE Adaptive Packet Video Streaming Over IP Networks - LaBRI

classification layer which classifies the object between them. As we said, the result of the
classification is a set of AVOs sorted according to their importance in the scene. The classification
layer affects a final relative priority score (RPS) to each Access Unit to apply a differentiation
mechanism. This priority score reflects both the priority of particular AVO in the scene and the
priority of a single frame type (I, P, B or hierarchical stream if any BL or EL). In the rest of this
Section, we will interest to the prototype implementation which is based on a neural network
algorithm.
n
Let T = {(X,c)} be a training set of N labeled vectors, where Χ ∈R is a feature vector X =
(x 1 ,x 2 ,…,x n ) t and c ∈ T is its class label from an index set T (X is an AVO, and T is the set of
available classes (e.g. Diffserv PHB), in our context) . The variables x i are referred to as attributes
(e.g. QoS features of each AVO). A class is modeled by one or more prototype, which have
{ U , U , 2
K }
U m
1
as features.
A classifier is a mapping function called C defined as C: R n T, which assigns a class label in
T to each vector in R n (a vector is an MPEG-4 AVO features). Typically, the classifier is
represented by a set of model parameters Λ = { ∆ k
}. The classifier specifies a partitioning of the
feature space into regions R j = { X∈ R n n
: C(X) = j} where U R
j
≡ R and I R
j
≡ Φ .
It also induces a corresponding partitioning of the training set into subset T.
There are different methods based on this definition, which allow an automatic classification
of Vectors. We present in what below the classification method that used Radial Basis Function
(RBF) classification.
Figure 4-3 shows the RBF classifier. A vector to be classified is passed to a set of basis
functions, each returning one scale value ϕ j j=1, 2,…,l. The concrete choice of the basis function is
not critical; the common implementations prefer the radial basic function (the Gaussian “bell”
functions). Hence the name: RBF network, with :
j
j
ϕ (x) = e
j
2
D
−
2
2σ j
ϕ 1
+
F 1 (x)
X
ϕ 2
+
F 2 (x)
Choose
Largest
F j (x)
j
ϕ l
λ lk
+
F k (x)
Figure 4-3: RBF architecture for classification
68