MATH 414 Lecture 27

From Notes
Jump to navigation Jump to search

« previous | Wednesday, March 26, 2014 | next »


Wavelets

Haar scaling function: ϕ(x)={1x∈[0,1)0otherwise

Shifts: ϕ(x−k)

"Sampling" spaces

  • Integers: V0={f∈L2∣f is constant on k≤x<k+1,k∈ℤ} (sampling interval has unit length 1)
  • Half-Integers: V1={f∈L2∣f is constant on k2≤x<k+12} (sampling interval has length 12)
  • Vj={f∈L2∣f is constant on k2j≤x<k+12j} (sampling interval has length 2−j)
Note: As j increases, the "sampling" occurs at finer scales.


Properties

The following 5 properties institute a multi-resolution analysis.

Definition: The Vj's are called approximation spaces (or scaling spaces)

Nesting

Theorem. [Nesting.] The spaces Vj are nested:

…⊂Vj−1⊂Vj⊂Vj+1⊂…

Proof by induction. V0⊂V1.

If f∈V, then f is constant on 2−jk≤x<2−j(k+1). This implies that f is constant on 2−(j+1)k≤x<2−(j+1)(k+1) and on 2−(j+1)(k+1)≤x<2−(j+1)(k+2). Hence Vj⊂Vj+1

The hypothesis follows by induction on j.

quod erat demonstrandum


Density

One can approximate any f in L2 arbitrarily well by functions in Vj if j is large enough.

⋃j∈ℤVj=L2


Separation

⋂j∈ℤVj={0}

Orthonormal Basis

Theorem. The set {2j2ϕ(2jx−k)}k=−∞∞ is an orthonormal basis for Vj

Proof. We know from last time that ⟨ϕ(x−k),ϕ(x−ℓ)⟩=δk,ℓ

∫−∞∞2j2ϕ(2jx−k)ϕ(2j−ℓ)⋅2j2dx=2j∫−∞∞ϕ(2jx−k)ϕ(2jx−ℓ)dx

If we let u=2jx, then

2j∫−∞∞ϕ(2jx−k)ϕ(2jx−ℓ)dx=∫−∞∞ϕ(u−k)ϕ(u−ℓ)du=δk,ℓ
quod erat demonstrandum


Scaling Property

f(x)∈Vj if and only if f(2−jx)∈V0



We know if f∈Vj, then f∈Vj+1 (nesting).

f=∑k=−∞∞⟨2j2ϕ(2j(⋅)−k),f⟩L22j2ϕ(2jx−k)

Let akj=2j2⟨ϕ(2j(⋅)−k),f⟩L2

akj=2j∫−∞∞ϕ(2jx−k)f(x)dx=2j∫2−jk2−j(k+1)f(x)dx=12−j∫2−jk2−j(k+1)f(x)dx

This integral represents the average value [1] over the interval [2−jk,2−j(k+1))


Footnotes

  1. ↑ Average value of function ⟨f⟩ given by ⟨f⟩=1b−a∫abf(x)dx.