Haar wavelet

Uses your system voice; voice and rate in settings.

Snapshot

≈ 14 min read

Names in other languages

No names in other languages are held for this entry yet.

Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.

In mathematics, the Haar wavelet is a sequence of rescaled "square-shaped" functions which together form a wavelet family or basis. Wavelet analysis is similar to Fourier analysis in that it allows a target function over an interval to be represented in terms of an orthonormal basis.

The Haar sequence is now recognised as the first known wavelet basis and is extensively used as a teaching example. The Haar sequence was proposed in 1909 by Alfréd Haar. Haar used these functions to give an example of an orthonormal system for the space of square-integrable functions on the unit interval [0, 1].

The study of wavelets, and even the term "wavelet", did not come until much later. As a special case of the Daubechies wavelet, the Haar wavelet is also known as Db1. The Haar wavelet is also the simplest possible wavelet.

The technical disadvantage of the Haar wavelet is that it is not continuous, and therefore not differentiable. This property can, however, be an advantage for the analysis of signals with sudden transitions (discrete signals), such as monitoring of tool failure in machines.

The Haar wavelet's mother wavelet function \psi(t) can be described as : \psi(t) = \begin{cases} 1 \quad & 0 \leq t Its scaling function \varphi(t) can be described as : \varphi(t) = \begin{cases}1 \quad & 0 \leq t

In mathematics, the Haar wavelet is a sequence of rescaled "square-shaped" functions which together form a wavelet family or basis. Wavelet analysis is similar to Fourier analysis in that it allows a target function over an interval to be represented in terms of an orthonormal basis.

The Haar sequence is now recognised as the first known wavelet basis and is extensively used as a teaching example. The Haar sequence was proposed in 1909 by Alfréd Haar. Haar used these functions to give an example of an orthonormal system for the space of square-integrable functions on the unit interval [0, 1].

The study of wavelets, and even the term "wavelet", did not come until much later. As a special case of the Daubechies wavelet, the Haar wavelet is also known as Db1. The Haar wavelet is also the simplest possible wavelet.

The technical disadvantage of the Haar wavelet is that it is not continuous, and therefore not differentiable. This property can, however, be an advantage for the analysis of signals with sudden transitions (discrete signals), such as monitoring of tool failure in machines.

The Haar wavelet's mother wavelet function \psi(t) can be described as \psi(t) = \begin{cases} 1 \quad & 0 \leq t Its scaling function \varphi(t) can be described as \varphi(t) = \begin{cases}1 \quad & 0 \leq t \psi_{n,k}(t) = 2^{n / 2} \psi(2^n t-k), \quad t \in \mathbb{R}. This function is supported on the right-open interval , i.e., it vanishes outside that interval.

It has integral 0 and norm 1 in the Hilbert space L 2 ( \mathbb{R} ), \int_{\mathbb{R}} \psi_{n, k}(t) \, d t = 0, \quad \|\psi_{n, k}\|^2_{L^2(\mathbb{R})} = \int_{\mathbb{R}} \psi_{n, k}(t)^2 \, d t = 1. The Haar functions are pairwise orthogonal, \int_{\mathbb{R}} \psi_{n_1, k_1}(t) \psi_{n_2, k_2}(t) \, d t = \delta_{n_1n_2} \delta_{k_1k_2}, where \delta_{ij} represents the Kronecker delta.

Here is the reason for orthogonality: when the two supporting intervals I_{n_1, k_1} and I_{n_2, k_2} are not equal, then they are either disjoint, or else the smaller of the two supports, say I_{n_1, k_1} , is contained in the lower or in the upper half of the other interval, on which the function \psi_{n_2, k_2} remains constant. It follows in this case that the product of these two Haar functions is a multiple of the first Haar function, hence the product has integral 0.

The Haar system on the real line is the set of functions \{ \psi_{n,k}(t) \; : \; n \in \mathbb{Z}, \; k \in \mathbb{Z} \}. It is complete in L 2 ( \mathbb{R} ): The Haar system on the line is an orthonormal basis in L 2 ( \mathbb{R} ).

Haar wavelet properties

The Haar wavelet has several notable properties: r_n(t) = 2^{-n/2} \sum_{k=0}^{2^n - 1} \psi_{n, k}(t), \quad t \in [0, 1], \ n \ge 0. Notice that |r n (t)| = 1 on [0, 1). This is an orthonormal system but it is not complete.

In the language of probability theory, the Rademacher sequence is an instance of a sequence of independent Bernoulli random variables with mean 0. The Khintchine inequality expresses the fact that in all the spaces L p ([0, 1]), , the Rademacher sequence is equivalent to the unit vector basis in ℓ 2 .

In particular, the closed linear span of the Rademacher sequence in L p ([0, 1]), , is isomorphic to ℓ 2 . is the family of continuous functions on [0, 1] consisting of the constant function 1, and of multiples of indefinite integrals of the functions in the Haar system on [0, 1], chosen to have norm 1 in the maximum norm. This system begins with s 0  = 1, then is the indefinite integral vanishing at 0 of the function 1, first element of the Haar system on [0, 1].

Next, for every integer , functions are defined by the formula s_{n, k}(t) = 2^{1 + n/2} \int_0^t \psi_{n, k}(u) \, d u, \quad t \in [0, 1], \ 0 \le k These functions are continuous, piecewise linear, supported by the interval that also supports . The function is equal to 1 at the midpoint of the interval  , linear on both halves of that interval.

It takes values between 0 and 1 everywhere. The Faber–Schauder system is a Schauder basis for the space C([0, 1]) of continuous functions on [0, 1].

For every f in C([0, 1]), the partial sum f_{n+1} = a_0 s_0 + a_1 s_1 + \sum_{m = 0}^{n-1} \Bigl( \sum_{k=0}^{2^m - 1} a_{m,k} s_{m, k} \Bigr) \in C([0, 1]) of the series expansion of f in the Faber–Schauder system is the continuous piecewise linear function that agrees with f at the points , where . Next, the formula f_{n+2} - f_{n+1} = \sum_{k=0}^{2^n - 1} \bigl( f(x_{n,k}) - f_{n+1}(x_{n, k}) \bigr) s_{n, k} = \sum_{k=0}^{2^n - 1} a_{n, k} s_{n, k} gives a way to compute the expansion of f step by step.

Since f is uniformly continuous, the sequence {f n } converges uniformly to f. It follows that the Faber–Schauder series expansion of f converges in C([0, 1]), and the sum of this series is equal to f. \left\{ f : x \in [0, \pi] \rightarrow \sum_{n=0}^\infty a_n \cos(n x) \right\} \longrightarrow \left\{ T(f) : z \rightarrow \sum_{n=0}^\infty a_n z^n, \quad |z| \le 1 \right\}. Bočkarev's basis for A(D) is formed by the images under T of the functions in the Franklin system on [0, π].

Bočkarev's equivalent description for the mapping T starts by extending f to an even Lipschitz function g 1 on [−π, π], identified with a Lipschitz function on the unit circle T. Next, let g 2 be the conjugate function of g 1 , and define T(f) to be the function in A(D) whose value on the boundary T of D is equal to  . When dealing with 1-periodic continuous functions, or rather with continuous functions f on [0, 1] such that , one removes the function from the Faber–Schauder system, in order to obtain the periodic Faber–Schauder system.

The periodic Franklin system is obtained by orthonormalization from the periodic Faber–-Schauder system. One can prove Bočkarev's result on A(D) by proving that the periodic Franklin system on [0, 2π] is a basis for a Banach space A r isomorphic to A(D). The space A r consists of complex continuous functions on the unit circle T whose conjugate function is also continuous.

Haar matrix

The 2×2 Haar matrix that is associated with the Haar wavelet is H_2 = \begin{bmatrix} 1 & 1 \\ 1 & -1 \end{bmatrix}. Using the discrete wavelet transform, one can transform any sequence (a_0,a_1,\dots,a_{2n},a_{2n+1}) of even length into a sequence of two-component-vectors \left(\left(a_0,a_1\right),\left(a_2,a_3\right),\dots,\left(a_{2n},a_{2n+1}\right)\right) .

If one right-multiplies each vector with the matrix H_2 , one gets the result \left(\left(s_0,d_0\right),\dots,\left(s_n,d_n\right)\right) of one stage of the fast Haar-wavelet transform. Usually one separates the sequences s and d and continues with transforming the sequence s. Sequence s is often referred to as the averages part, whereas d is known as the details part.

If one has a sequence of length a multiple of four, one can build blocks of 4 elements and transform them in a similar manner with the 4×4 Haar matrix H_4 = \begin{bmatrix} 1 & 1 & 1 & 1 \\ 1 & 1 & -1 & -1 \\ 1 & -1 & 0 & 0\\ 0 & 0 & 1 & -1 \end{bmatrix}, which combines two stages of the fast Haar-wavelet transform. Compare with a Walsh matrix, which is a non-localized 1/–1 matrix.

Generally, the 2N×2N Haar matrix can be derived by the following equation. H_{2N} = \begin{bmatrix} H_{N} \otimes [1, 1] \\ I_{N} \otimes [1, -1] \end{bmatrix} where I_{N} = \begin{bmatrix} 1 & 0 & \dots & 0 \\ 0 & 1 & \dots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dots & 1 \end{bmatrix} and \otimes is the Kronecker product.

The Kronecker product of A \otimes B , where A is an m×n matrix and B is a p×q matrix, is expressed as A \otimes B = \begin{bmatrix} a_{11}B & \dots & a_{1n}B \\ \vdots & \ddots & \vdots \\ a_{m1}B & \dots & a_{mn}B\end{bmatrix}. An un-normalized 8-point Haar matrix H_8 is shown below H_{8} = \begin{bmatrix} 1&1&1&1&1&1&1&1 \\ 1&1&1&1&-1&-1&-1&-1 \\ 1&1&-1&-1&0&0&0&0& \\ 0&0&0&0&1&1&-1&-1 \\ 1&-1&0&0&0&0&0&0& \\ 0&0&1&-1&0&0&0&0 \\ 0&0&0&0&1&-1&0&0& \\ 0&0&0&0&0&0&1&-1 \end{bmatrix}.

Note that, the above matrix is an un-normalized Haar matrix. The Haar matrix required by the Haar transform should be normalized. From the definition of the Haar matrix H , one can observe that, unlike the Fourier transform, H has only real elements (i.e., 1, -1 or 0) and is non-symmetric.

Take the 8-point Haar matrix H_8 as an example. The first row of H_8 measures the average value, and the second row of H_8 measures a low frequency component of the input vector. The next two rows are sensitive to the first and second half of the input vector respectively, which corresponds to moderate frequency components.

The remaining four rows are sensitive to the four section of the input vector, which corresponds to high frequency components.

Haar transform

The Haar transform is the simplest of the wavelet transforms. This transform cross-multiplies a function against the Haar wavelet with various shifts and stretches, like the Fourier transform cross-multiplies a function against a sine wave with two phases and many stretches.

H_4 = \frac{1}{2} \begin{bmatrix} 1 & 1 & 1 & 1 \\ 1 & 1 & -1 & -1 \\ \sqrt{2} & -\sqrt{2} & 0 & 0 \\ 0 & 0 & \sqrt{2} & -\sqrt{2}\end{bmatrix} The Haar transform can be thought of as a sampling process in which rows of the transformation matrix act as samples of finer and finer resolution. Compare with the Walsh transform, which is also 1/–1, but is non-localized.

Input and output length are the same. However, the length should be a power of 2, i.e. N = 2^k, k\in \mathbb{N} . It can be used to analyse the localized feature of signals.

Due to the orthogonal property of the Haar function, the frequency components of input signal can be analyzed. y_n = H_n x_n The Haar transform matrix is real and orthogonal. Thus, the inverse Haar transform can be derived by the following equations. H = H^*, H^{-1} = H^T, \text{ i.e. } HH^T = I where I is the identity matrix.

For example, when n = 4 H_4^{T}H_4 = \frac{1}{2}\begin{bmatrix} 1&1&\sqrt{2}&0 \\ 1&1&-\sqrt{2}&0 \\ 1&-1&0&\sqrt{2} \\ 1&-1&0&-\sqrt{2}\end{bmatrix} \cdot\; \frac{1}{2}\begin{bmatrix} 1&1&1&1 \\ 1&1&-1&-1 \\ \sqrt{2}&-\sqrt{2}&0&0 \\ 0&0&\sqrt{2}&-\sqrt{2}\end{bmatrix} The Haar transform coefficients of a n=4-point signal x_{4} = [1,2,3,4]^{T} can be found as y_{4} = H_4 x_4 = \frac{1}{2}\begin{bmatrix} 1&1&1&1 \\ 1&1&-1&-1 \\ \sqrt{2}&-\sqrt{2}&0&0 \\ 0&0&\sqrt{2}&-\sqrt{2}\end{bmatrix} \begin{bmatrix} 1 \\ 2 \\ 3 \\ 4\end{bmatrix} = \begin{bmatrix} 5 \\ -2 \\ -1/\sqrt{2} \\ -1/\sqrt{2}\end{bmatrix} The input signal can then be perfectly reconstructed by the inverse Haar transform \hat{x_{4}} = H_{4}^{T}y_{4} = \frac{1}{2}\begin{bmatrix} 1&1&\sqrt{2}&0 \\ 1&1&-\sqrt{2}&0 \\ 1&-1&0&\sqrt{2} \\ 1&-1&0&-\sqrt{2}\end{bmatrix} \begin{bmatrix} 5 \\ -2 \\ -1/\sqrt{2} \\ -1/\sqrt{2}\end{bmatrix} = \begin{bmatrix} 1 \\ 2 \\ 3 \\ 4 \end{bmatrix}

See also

Dimension reduction Walsh matrix Walsh transform Wavelet Chirplet Signal Haar-like feature Strömberg wavelet Dyadic transformation

References

Charles K. Chui, An Introduction to Wavelets, (1992), Academic Press, San Diego, English Translation of Haar's seminal article:

[https://web.archive.org/web/20071109204233/http://www.tomgibara.com/computer-vision/haar-wavelet Free Haar wavelet filtering implementation and interactive demo Free Haar wavelet denoising and lossy signal compression

Text derived from https://en.wikipedia.org/w/index.php?oldid=1298249358 under CC BY-SA 4.0; modifications have been made.

Sources cited in this revision

5 named source(s) extracted from the cited revision. A source with no author is shown by its publication; nothing is invented.