Multi-Resolution Analysis
📂Fourier AnalysisMulti-Resolution Analysis
Definition
If a sequence of closed subspaces L2(R) and a function ϕ∈V0 satisfy the following conditions, then ({Vj},ϕ) is called a multiresolution analysis.
(a) For each Vj, ⋯V−1⊂V0⊂V1⋯ holds.
(b) ∪j∈ZVj=L2(R) and ∩j∈ZVj={0}.
(c) ∀j∈Z, Vj+1=D(Vj).
(d) If ∀k∈Z, f∈V0, then Tkf∈V0.
(e) {Tkϕ}k∈Z is an orthonormal basis of V0.
If ({Vj},ϕ) is a multiresolution analysis, then ϕ is said to generate the multiresolution analysis. Tk is a translation, D is a dilation.
Explanation
Condition (b) means that ∪jVj is dense in L2(R), meaning that there exists an approximation g∈∪jVj by some f∈L2(R). If such g belongs to VJ, then according to condition (a), g is contained in all Vj where j≥J. Moreover, from the definition, the following facts are established.
Theorem
If conditions (c) and (d) of the definition are satisfied, then the following two facts also hold for all j∈Z:
(f) Vj=Dj(V0)
(g) Vj=span{DjTkϕ}
Proof
(f)
Assuming (c) holds, then for all j∈N
Vj=D(Vj−1)=DD(Vj−2)=⋯=DjV(0)
Also, for all j∈{−1,−2,⋯}
Vj=D−1(Vj+1)=D−1D−1(Vj+2)=⋯=(D−1)−j(V0)=Dj(V0)
■
(g)
Since {Tkϕ}k∈Z is an orthonormal basis of V0
V0=span{Tkϕ}k∈Z
is established. Hence, by (f)
Vj=Dj(V0)=Dj(span{Tkϕ}k∈Z)=span{DjTkϕ}k∈Z
■