Skip to main content

An efficient computation of discrete orthogonal moments for bio-signals reconstruction

Abstract

Bio-signals are extensively used in diagnosing many diseases in wearable devices. In signal processing, signal reconstruction is one of the essential applications. Discrete orthogonal moments (DOMs) are effective analysis tools for signals that can extract digital information without redundancy. The propagation of numerical errors is a significant challenge for the computation of DOMs at high orders. This problem damages the orthogonality property of these moments, which restricts the ability to recover the signal's distinct and unique components with no redundant information. This paper proposes a stable computation of DOMs based on QR decomposition methods: the Gram–Schmidt, Householder, and Given Rotations methods. It also presents a comparative study on the performance of the types of moments: Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moments. The proposed algorithm's evaluation is done using the MIT-BIH arrhythmia dataset in terms of mean square error and peak signal to noise ratio. The results demonstrate the superiority of the proposed method in computing DOMs, especially at high moment orders. Moreover, the results indicate that the Householder method outperforms Gram–Schmidt and Given Rotations methods in execution time and reconstruction quality. The comparative results show that Tchebichef, Krawtchouk, and Charlier moments have superior reconstruction quality than Hahn and Meixner moments, and Tchebichef generally has the highest performance in signal reconstruction.

Introduction

With the explosive growth in computer technology and signal analysis tools, computer-aided analysis of bio-signals has become a common part of clinical. Bio-signal reconstruction is one of the important applications in bio-signal processing. The study of the literature on image analysis techniques indicates that the method of orthogonal moments plays a significant role in each of its important fields. These fields include image reconstruction [1, 2], face recognition [3], image classification [4, 5], image watermarking [6], image encryption [7], image compression [8, 9], color stereo image analysis [10]. Orthogonal moments are classified as continuous or discrete depending on whether the kernel functions are orthogonal in the continuous or discrete domain. Continuous orthogonal can be utilized to characterize an image with minimal redundant information. But even so, computing these moments needs a coordinate transformation and an estimate of the continuous moment's integrals. This adds computational complexity and introduces approximation errors [11]. To this end, many researchers have started to use discrete orthogonal moments [12, 13]. Zhu et al. [14, 15] demonstrated that discrete orthogonal moments are more effective than continuous orthogonal moments at representing images. The types of discrete orthogonal moments (DOMs) according to their corresponding discrete orthogonal polynomials include Tchebichef [16, 17], Krawtchouk[18,19,20,21], Charlier [22,23,24], Hahn [25, 26] and Meixner [27, 28] moments. At the present time, Discrete Orthogonal Moments (DOMs) are gaining popularity in analyzing one-dimensional signals due to their effectiveness in capturing digital information without redundancy. In order to compute DOMs, we have to compute Kernel discrete orthogonal polynomials (DOPs).

The computation of high-order DOPs faces a major problem which is the propagation of numerical errors. This problem destroys the orthogonality property of these polynomials, which affects the ability to extract the signal's distinct and unique components with no information redundancy. To address this problem, we propose using QR decomposition methods to maintain the orthogonality property by re-orthonormalization DOPs. There are many ways for QR decomposition, like the Gram–Schmidt method, the Householder method, and the Given Rotations method [29]. These methods maintain the high-order DOPs orthogonality property effectively. Thus, using the DOMs to analyze large-size signals will become highly efficient Due to significant improvements in the computation of DOPs. Our paper presents several contributions that can be summarized as follows:

  • Testing Discrete Orthogonal Moments (DOMs) in bio-signals analysis and reconstruction.

  • Proposing a new modified version of DOPs by the QR decomposition methods like the Gram–Schmidt, Householder, and Given Rotations methods. In addition to comparing methods of QR decomposition to estimate the best methods.

  • Presenting comparative study between the different types of moments to estimate the best moment in analyzing and reconstructing bio-signals.

The rest of the paper is structured into five sections: Sect. 2 outlines the Recursive relation of Discrete Orthogonal Polynomials (DOPs). Discrete Orthogonal Moments (DOMs) will be discussed in Sect. 3. Section 4 shows the proposed procedure for ensuring the orthogonality property of discrete polynomials. Experimental results and discussion are presented in Sect. 5. Discussions are shown in Sect. 6. In Sect. 7, we conclude our work.

Recursive relation of discrete orthogonal polynomials (DOPs)

The discrete orthogonal polynomials are the polynomial solutions of the given difference equation

$$\sigma (x)\Delta \nabla {p}_{n}(x)+\tau (x)\Delta {p}_{n}(x)+{\lambda }_{n}{p}_{n}(x)=0$$
(1.a)

where \(\Delta {p}_{n}\left(x\right)={p}_{n}\left(x+1\right)-{p}_{n}\left(x\right) and \nabla {p}_{n}(x)={p}_{n}(x)-{p}_{n}(x-1)\) indicates backward finite-difference operator and forward finite difference operator, respectively. \(\sigma (x)\text{ and }\tau (x)\) denote first and second degree functions. \({\lambda }_{n}\) indicate a suitable constant.

The polynomials \({p}_{n}(x)\) satisfy an orthogonality relation of the form

$$\sum\limits_{{x = 0}}^{s} {p_{n} } \left( x \right)p_{m} \left( x \right)w\left( x \right) = d_{n}^{2} \cdot \delta _{{mn}} ,\quad 0 \le m,n \le s$$
(1.b)

where \(w\left(x\right)\) is the weight function, \({d}_{n}^{2}\) denotes the square of the norm of the corresponding orthogonal polynomials and \({\delta }_{mn}\) denotes the Dirac function. The normalized orthogonal polynomials can be obtained by utilizing the square norm and weighted function

$${\widetilde{p}}_{n}(x)={p}_{n}(x)\sqrt{\frac{w(x)}{{d}_{n}^{2}}}, \quad n={0,1},\dots ,s$$
(1.c)

Therefore, the orthogonal property of normalized orthogonal polynomials in (1b) can be rewritten as

$$\sum_{x=0}^{s} {\widetilde{p}}_{m}(x){\widetilde{p}}_{n}(x)={\delta }_{mn}, \quad 0\le m,n\le s$$
(1.d)

A general formula for getting the normalized discrete orthogonal polynomials \({\widetilde{p}}_{n}(x)\) of order \(n\) is defined three-term recursive relation as follows [11]:

$${\widetilde{p}}_{n}\left(x\right)=\left(\frac{B*D}{A}\right){\widetilde{p}}_{n-1}\left(x\right)+\left(\frac{C*E}{A}\right){\widetilde{p}}_{n-2}\left(x\right)$$
(1.e)

where A, B, C, D, and E are coefficients independent of each polynomial set shown in Table 1. These discrete orthogonal polynomials include Tchebichef\({{t}}_{{n}}\left({x};{ N}\right)\), Krawtchouk \({k}_{n}\left(x;P,N\right)\), Charlier \({c}_{n}(x)\), Hahn \({h}_{n}^{(\alpha ,\beta )}(x)\) and Meixner \({m}_{n}^{(a,b)}(x)\) Polynomials.\({\widetilde{p}}_{n-1}(x)\) and \({\widetilde{p}}_{n-2}(x)\) are the zero-order and first-order polynomials, respectively.

Table 1 Values of A, B, C, D, and E for each of the polynomial set Tchebichef \({{t}}_{{n}}\left({x};{ N}\right)\), Krawtchouk \({k}_{n}\left(x;P,N\right)\), Charlier \({c}_{n}\left(x\right)\), Hahn \({h}_{n}^{\left(\alpha ,\beta \right)}\left(x\right)\) and Meixner \({m}_{n}^{\left(a,b\right)}\left(x\right)\)

Tchebichef polynomials

The nth Tchebichef polynomials \(t_{n} \left( x \right)\) are defined by hypergeometric function as the follows

$$t_{n} \left( x \right) = (1 - N)_{n3} F_{2} \left( { - n, - x,1 + n;1,1 - N;1} \right),\; n,x = 0,1,2, \ldots ,N - 1.$$
(2)

From Eq. (1.e) and Table 1, we obtain the recursive relation of discrete orthogonal Tchebichef polynomials as follows:

$${t}_{n}(x)=\left({\beta }_{1}x+{\beta }_{2}\right){t}_{n-1}(x)+{\beta }_{3}{t}_{n-2}(x)$$
(3)

with

$$\begin{aligned} t_{0} \left( x \right) & = \frac{1}{{\sqrt N }}, \\ t_{1} \left( x \right) & = \left( {2x + 1 - N} \right)\sqrt {\frac{3}{{N\left( {N^{2} - 1} \right)}}} . \\ \end{aligned}$$
(4)

where

$$\begin{aligned} \beta _{1} & = \frac{2}{n}\sqrt {\frac{{4n^{2} - 1}}{{N^{2} - n^{2} }}} \\ \beta _{2} & = \frac{{1 - N}}{n}\sqrt {\frac{{4n^{{2 - 1}} }}{{N^{2} - n^{2} }},} \\ \beta _{3} & = \frac{{1 - n}}{n}\sqrt {\frac{{2n + 1}}{{2n - 3}}} \sqrt {\frac{{N^{2} - (n - 1)^{2} }}{{N^{2} - n^{2} }}.} \\ \end{aligned}$$
(5)

Krawtchouk polynomials

Krawtchouk polynomials \(k_{n} \left( {x,p} \right)\) of order n are defined by hypergeometric function as the follows

$$k_{n} \left( {x,p} \right) =_{ 2} F_{1} \left( { - n, - x; - N;\frac{1}{p}} \right) ,\quad n,x = 0,1,2, \ldots ,N - 1.$$
(6)

The recursive relation of discrete orthogonal Krawtchouk polynomials can be calculated using Eq. (1.e) and Table 1 as:

$$k_{n} \left( {x,p} \right) = \left( {x + \beta_{1} } \right)\beta_{2} k_{n - 1} \left( {x,p} \right) - \beta_{3} k_{n - 2} \left( {x,p} \right)$$
(7)

with

$$\begin{aligned} k_{0} \left( {x,p} \right) & = \sqrt {\frac{{N!p^{x} (1 - p)^{{N - x}} }}{{x!\left( {N - x} \right)!}}} , \\ k_{1} \left( {x,p} \right) & = ( - p\left( {N - x} \right) + x(1 - p)) \times \sqrt {\frac{{(N - 1)!p^{{x - 1}} (1 - p)^{{N - x - 1}} }}{{x!\left( {N - x} \right)!}}} \\ \end{aligned}$$
(8)

where

$$\begin{aligned} \beta _{1} & = \left( {1 - n - p\,\left( {N - 2n + 2} \right)} \right), \\ \beta _{2} & = \sqrt {\frac{1}{{p(1 - p)(N - n + 1)n}}} \\ \beta _{3} & = \sqrt {\frac{{\left( {N - n + 2} \right)\left( {n - 1} \right)}}{{\left( {N - n + 1} \right)n}}} ,\quad 0 < p < 1. \\ \end{aligned}$$
(9)

Charlier polynomials

Charlier polynomials \({C}_{n}^{{a}_{1}}(x)\) of order n are defined by hypergeometric function as the follows

$$c_{n}^{{a_{1} }} \left( x \right) =_{2} F_{0} \left( { - n, - x;; - 1/a_{1} } \right),\quad n,x = 0,1,2, \ldots ,N - 1 \;{\text{and}}\;a_{1} > 0$$
(10)

By substituting coefficients of A, B, C, D, and E from Table 1 in Eq. (1.e), we conclude the recursive relation of discrete orthogonal Charlier polynomials as follows:

$$C_{n}^{{a_{1} }} \left( x \right) = \left( {\beta_{1} - x} \right) \beta_{2} C_{n - 1}^{{a_{1} }} \left( x \right) - \beta_{3} C_{n - 2}^{{a_{1} }} \left( x \right)$$
(11)

with

$$\begin{aligned} c_{0}^{{a_{1} }} \left( x \right) & = \sqrt {\frac{{e^{{ - a_{1} }} a_{1}^{x} }}{{x!}},} \\ C_{1}^{{a_{1} }} (x) & = \frac{{a_{1} - x}}{{a_{1} }}\sqrt {\frac{{e^{{ - a_{1} }} a_{1}^{{x + 1}} }}{{x!}}} \\ \end{aligned}$$
(12)

where

$$\begin{aligned} \beta _{1} & = \left( {a_{1} + n - 1} \right), \\ \beta _{2} & = \sqrt {\frac{1}{{na_{1} }}} , \\ \beta _{3} & = \sqrt {\frac{{n - 1}}{n}} . \\ \end{aligned}$$
(13)

Hahn polynomials

The nth Hahn polynomials \({h}_{n}^{\left(\alpha ,\beta \right)}(x)\) are defined by hypergeometric function as the follows.

$$h_{n}^{{\left( {\alpha ,\beta } \right)}} \left( x \right) = \frac{{( - 1)^{n} (\beta + 1)_{n} (N - n)_{n} }}{{n!}} \times _{3} F_{2} \left( { - n, - x,n + 1 + \alpha + \beta ;\beta + {\text{1}},{\text{1}} - N;1} \right),\quad n,x = {\text{0}},{\text{1}},2, \ldots ,N - 1.$$
(14)

By substituting coefficients of A, B, C, D, and E from Table 1 in Eq. (1.e), we conclude the recursive relation of discrete orthogonal Charlier polynomials as follows:

$$h_{n}^{{\left( {\alpha ,\beta } \right)}} \left( x \right) = \left( {x - \beta_{1} } \right)\beta_{2} h_{n - 1}^{{\left( {\alpha ,\beta } \right)}} \left( x \right) - \beta_{3} h_{n - 2}^{{\left( {\alpha ,\beta } \right)}} \left( x \right)$$
(15)

with

$$\begin{aligned} h_{0}^{{\left( {\alpha ,\beta } \right)}} \left( x \right) & = \sqrt {\frac{{(\alpha + 1)_{\beta } (\alpha + \beta + 1)}}{{(N - \alpha )_{{\beta + 1}} }}} \\ h_{1}^{{\left( {\alpha ,\beta } \right)}} \left( x \right) & = \left( {\alpha + \beta + 2} \right)x - \left( {\beta + 1} \right)\left( {N - 1} \right) \times \sqrt {\frac{{\alpha + \beta + 3}}{{\left( {\alpha + 1} \right)\left( {\beta + 1} \right)\left( {n - 1} \right)\left( {N + 1 + \beta + 1} \right)}}} . \\ \end{aligned}$$
(16)

where

$$\begin{aligned} \beta _{1} & = \frac{{\alpha - \beta + 2N - 2}}{4} + \frac{{\left( {\beta ^{2} - \alpha ^{2} } \right)(\alpha + \beta + 2N)}}{{4(\alpha + \beta + 2n - 2)(\alpha + \beta + 2n)}}, \\ \beta _{2} & = \sqrt {\frac{{\left( {\alpha + \beta + 2n} \right)^{4} - \left( {\alpha + \beta + 2n} \right)^{2} }}{{n(N - n)\left( {\alpha + n} \right)\left( {\beta + n} \right)\left( {\alpha + \beta + n + N} \right)(\alpha + \beta + n)}}} , \\ \beta _{3} & = \frac{{\alpha + \beta + 2n}}{{\alpha + \beta + 2n - 2}} \times \sqrt {\frac{{(n - 1)(\alpha + n - 1)(\beta + n - 1)(N - n + 1)}}{{n(\alpha + n)(\beta + n)(N - n)(\alpha + \beta + n)}}} \quad \times \sqrt {\frac{{(\alpha + \beta + n - 1)(\alpha + \beta + 2n + 1)(\alpha + \beta + N + n - 1)}}{{(\alpha + \beta + 2n - 3)(\alpha + \beta + n + N)}}} . \\ \end{aligned}$$
(17)

\(\text{where }\,\alpha ,\beta >0\text{.}\)

Meixner polynomials

Meixner polynomials \({m}_{n}^{\left(a,b\right)}(x)\) of order n are defined by hypergeometric function as the follows

$$m_{n}^{{\left( {a,b} \right)}} \left( x \right) = (a)_{n2} F_{1} \left( { - n, - x;a;1 - 1/b} \right),\;n,x = 0,1,2, \ldots ,N - 1.$$
(18)

From Eq. (1.e) and Table 1, we obtain the recursive relation of discrete orthogonal Meixner polynomials as follows:

$$m_{n}^{{\left( {a,b} \right)}} \left( x \right) = \left( {x\beta_{1} + \beta_{2} } \right)m_{n - 1}^{{\left( {a,b} \right)}} \left( x \right) - \beta_{3} m_{n - 2}^{{\left( {a,b} \right)}} \left( x \right)$$
(19)

with

$$\begin{aligned} m_{0}^{{\left( {a,b} \right)}} \left( x \right) & = \sqrt {\frac{{b^{x} \left( {a + x - 1} \right)!}}{{x!\left( {a - 1} \right)!}}(1 - b)^{a} } , \\ m_{1}^{{\left( {a,b} \right)}} \left( x \right) & = \left( {a + x - \frac{x}{b}} \right) \times \sqrt {\frac{{b^{x} \left( {a + x - 1} \right)!}}{{x!\left( {a - 1} \right)!}}\frac{{b(1 - b)^{a} }}{a}} . \\ \end{aligned}$$
(20)

where

$$\begin{aligned} \beta _{1} & = \left( {b - 1} \right)\sqrt {\frac{1}{{n\left( {a + n - 1} \right)b}}} , \\ \beta _{2} & = \left( {n - 1 + bn - b + ab} \right)\sqrt {\frac{1}{{n\left( {a + n - 1} \right)b}}} , \\ \beta _{3} & = \sqrt {\frac{{\left( {n - 1} \right)\left( {n - 2 + a} \right)}}{{n\left( {a + n - 1} \right)}}} , \\ \end{aligned}$$
(21)

\(\text{where }\,0<b<1\text{ and }a>0\text{.}\)

Discrete orthogonal moments (DOMs)

The discrete orthogonal moments are a set of moments calculated by discrete orthogonal polynomials. The set of discrete orthogonal one-dimensional (1D) moments are defined as follows [11]:

$${M}_{n}=\sum_{x=0}^{N-1} {p}_{n}\left(x\right)s\left(x\right), \quad n={0,1}\dots ,N-1.$$
(22)

where \(s\left(x\right)\) is a one-dimensional signal of size \(1\times N,\) \({M}_{n}\) is a set of moment coefficients of the signal \(s\left(x\right)\) and \(p(x)\) is orthogonal polynomials of order n (Tchebichef \({{t}}_{{n}}\left({x};{ N}\right)\), Krawtchouk \({k}_{n}\left(x;P,N\right)\), Charlier \({c}_{n}\left(x\right)\), Hahn \({h}_{n}^{\left(\alpha ,\beta \right)}\left(x\right)\) and Meixner \({m}_{n}^{\left(a,b\right)}\left(x\right)\)).

The reconstructed signal \(S(x)\) is calculated from the inverse transformation of the orthogonal moment as follows:

$$S\left( x \right) = \mathop \sum \limits_{n = 0}^{N - 1} { }M_{n} p_{n} \left( x \right), \quad x = 0,1,2, \ldots ,N - 1$$
(23)

Using the following matrix form decreases the time and complexity of 1D orthogonal moment computations significantly:

$${M}_{n}=\left[\begin{array}{cccc}{p}_{0}(0)& {p}_{0}(1)& \dots & {p}_{0}(N-1)\\ {p}_{1}(0)& {p}_{1}(1)& \dots & {p}_{1}(N-1)\\ \vdots & \vdots & \vdots & \vdots \\ {p}_{n}(0)& {p}_{n}(1)& \dots & {p}_{n}\left(N-1\right)\end{array}\right]\times \left[\begin{array}{c}s(0)\\ s(1)\\ \vdots \\ s(N-1)\end{array}\right]$$
(24)

\(\text{where }\,{ M}_{n}\) indicates orthogonal polynomials of order n \(,s\text{ denotes } 1\times N \text{signal vector.}\)

Ensuring the orthogonality property of discrete polynomials

In this section, we propose a procedure for ensuring the orthogonality property of discrete polynomials. According to the orthogonality property, polynomials matrix (\({p}_{n}\left(x\right)\)) satisfies the following relation:

$$p_{n} \left( x \right)^{T} p_{n} \left( x \right) = I_{n}$$
(25)

where \({I}_{n}\) denotes the identity matrix.

To avoid numerical errors propagation and preserve the orthogonality property of DOPs, we present an efficient method for re-orthonormalizing \({p}_{n}(x)\) matrix columns using QR decomposition methods. In these methods, a matrix \(A=[ {u}_{1}\),\({u}_{2},\dots ,{u}_{n-1},{u}_{n}]\) of size \(n\times m\) factored as\(A=QR\), where Q is an \(n\times m\) matrix with orthogonal columns (\({Q}^{T}Q=I\)) and \(R\) is an \(m\times m\) upper triangular matrix [29]. In our situation, \(R\) matrix contains just recursive computation errors. The primary purpose of these ways is to generate the orthogonal \(Q(n\times m)\) matrix from \({p}_{n}(x)\) that contains round-off errors. Many ways are used in \(QR\) decomposition, such as the Gram–Schmidt method, the Householder method, and the Given Rotations method [30].

Computation DOPs with modified Gram–Schmidt method (MGSM)

One of the most common algorithms for applying QR decomposition is the Gram–Schmidt (GS) method. It is a simple procedure for generating an orthogonal or orthonormal basis for any nonzero \({R}^{n}\) subspace [31]. Although the modified Gram–Schmidt method is always preferred because it avoids potentially costly cancellation errors, it is not as good numerically as the Givens or Householder approaches [29]. Algorithm 1 summarizes the proposed implementation of DOPs using MGSM.

figure a

Computation DOPs with Householder method (HM)

The main way to apply QR decomposition is with the Householder method. [29]. This approach is regarded to be more numerically stable than the Gram–Schmidt orthogonalization method for QR matrix decomposition. The proposed computation of DOPs with HM is illustrated in Algorithm 2.

figure b

Computation DOPs with Given Rotations method (GRM)

The Given Rotations method is an alternative to the Modified Gram–Schmidt method and Householder method for calculating QR decomposition [29]. The Proposed Algorithm for computing DOPs using GRM is reported in Algorithm 3.

figure c

Results

The experiments of this study are performed on a personal computer using Matlab Software (version R2014a) on Microsoft Windows 7, 32-bit Edition, Intel Core i3 processor, and 4 GB RAM machine. Performance evaluation has been done by ECG signals from MIT-BIH arrhythmia dataset [32], which contain cardiac information from large numbers of patients. These recordings were obtained at a sampling frequency of 360 Hz (360 samples per second) with 11-bit resolution. Our numerical simulations are presented in three sections: the first is to evaluate the performance of the proposed re-orthonormalization methods in the quality of reconstruction signals. The second compares the three proposed re-orthogonalization methods (Gram–Schmidt, Householder, and Given Rotations) in signal reconstruction quality. The third is a comparative study on the performance of Discrete Orthogonal Moments in signal reconstruction. The quality of the reconstructed signal is evaluated based on the following criteria:

  • Peak signal to noise ratio (\(\mathrm{PSNR}\)):

    \(\mathrm{PSNR}\) is the highest possible signal power ratio to the corrupting noise power. It is presented as follows:

    $$\mathrm{PSNR}=20\times {{log}}_{10}\frac{max\left|s(x)\right|}{\sqrt{MSE}}$$
  • Mean-Square Error (MSE): the reconstruction error between the original and reconstructed signals.

    $$MSE=\frac{1}{N}\sum_{x=0}^{N-1} (s\left(x\right)-S\left(x\right){)}^{2}$$

    where \(s\left(x\right)\) and \(S\left(x\right)\) are the original signal and reconstructed signal, respectively.

In this experiment, the parameters of polynomials are set as \(p=0.5\) for Krawtchouk, \({a}_{1}=140\) for Charlier,\(\alpha ,\beta =100\) for Hahn, and \(a=512 ,b= 0.5\) for Meixner. The signal size is N = 3600, and the order of the DOMs used is 200.

Reconstruction quality of DOMs computed using the proposed re-orthonormalization method

We started by investigating the superiority of the proposed re-orthogonalization methods with the discrete orthogonal moments in reconstruction quality signals. As shown in Table 2, we test the Tchebichef moments with and without the Householder method as one of the proposed methods for reconstructing the signals. The results obtained in Table 2 show that using the Householder method significantly improves the reconstruction quality of all records used. Tchebichef moments with Householder provide a high Peak signal to noise ratio (PSNR) with very low Mean-Square Error (MSE) values compared to Tchebichef moments. The average reconstruction errors PSNR and MSE of the proposed methods are 109.147 and 0.0564, respectively, as reported in Table 2. Figure 1 presents the reconstructed signal's reconstruction errors (PSNR, MSE) using Tchebichef moments with and without Householder method. It confirms the superiority of the proposed methods in reconstructed signals. Figure 2 shows the reconstructed “Rec. 107” signal by Tchebichef moments with and without the Householder method.

Table 2 Comparison of reconstruction errors (PSNR and MSE) using Tchebichef moments with and without Householder method
Fig. 1
figure 1

The average values a PSNR and b MSE of the reconstructed signals using Tchebichef moments with and without Householder method

Fig. 2
figure 2

Set of reconstructed “Rec. 107” signal using Tchebichef moments with and without the Householder method

Comparison of reconstruction quality for the proposed re-orthonormalization methods

In the previous section, we investigated the ability of the proposed procedure to maintain the orthogonality property of the discrete polynomials in the reconstruction of the signal. There are three methods in the proposed procedure mentioned, and they are the Gram–Schmidt method (MGSM), the Householder method (HM), and the Given Rotations method (GRM). This section will investigate which of the three methods is preferable in signal reconstruction quality and execution time. We have used Tchebichef moments with the three proposed methods (MGSM, HM, and GRM) to reconstruct the signals and summarized the results in Table 3. Figure 3 also illustrates the reconstruction errors (PSNR, MSE) of the three proposed methods using Tchebichef moments. The results displayed in Table 3 and Fig. 3 demonstrate outperforming HM on MGSM and GRM in PSNR and MSE on all records used. The reconstructed “Rec. 115” signal by using Tchebichef moments with Gram–Schmidt, Householder, and Given Rotations methods are shown in Fig. 4.

Table 3 Comparison of reconstruction errors (PSNR and MSE) for re-orthogonalization methods (MGSM, HM, and GRM) using Tchebichef moments
Fig. 3
figure 3

The average values a PSNR and b MSE of the reconstructed signals using Tchebichef moments with Gram–Schmidt, Householder, and Given Rotations methods

Fig. 4
figure 4

Set of reconstructed “Rec. 115” signal using Tchebichef moments with Gram–Schmidt, Householder, and Given Rotations methods

We compare the execution time of HM on MGSM and GRM to discover which of the three methods is best in terms of execution time, as shown in Fig. 5. The visual inspection from Fig. 5 indicates that HM is faster than MGSM and GRM.

Fig. 5
figure 5

Average execution time using Tchebichef moments with re-orthonormalization methods (Gram–Schmidt, Householder, and Given Rotations)

Comparison of reconstruction quality for DOMs

This section determines which of the different types of moments is the best in the quality of the reconstructed signals. The compression performance of the discrete orthogonal moments (Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner) in signal reconstruction is presented in Table 4. In these experiments, the Householder method is used to preserve the orthogonality property methods in discrete orthogonal moments. Table 4 illustrates the resulting PSNR and MSE as reconstruction error metrics for 30 records from MIT-BIH arrhythmia dataset. The obtained results generally indicate that Tchebichef, Krawtchouk, and Charlier are superior to Hahn and Meixner in terms of PSNR and MSE. As for the three methods, Tchebichef, Krawtchouk, and Charlier, Tchebichef is relatively superior to Krawtchouk and Charlier. The average performance of the Tchebichef in terms of PSNR and MSE is 108.924 and 0.0580, respectively. Figures 6 and 7 depict the compression of the average PSNR and MSE of discrete orthogonal moments (Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner) in signal reconstruction. The reconstructed “Rec. 234” signal by using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder method is depicted in Fig. 8.

Table 4 Reconstructions errors (PSNR and MSE) of biosignal by Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moments with the Householder method
Fig. 6
figure 6

The average PSNR of the reconstructed signals using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder method

Fig. 7
figure 7

The average MSE of the reconstructed signals using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder method

Fig. 8
figure 8

Set of reconstructed “Rec. 234” signal using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder method

To further validate the efficiency of DOMs, reconstruction is conducted using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner with orders ranging between 50 and 200. Table 5 compares the quality of the signals reconstructed for the five moments in terms of PSNR and MSE in different moment orders. Figures 9 and 10 depict the curves of PSNR and MSE values corresponding to the reconstructed MIT-BIH Rec. 101 in different moments, respectively. As can be seen from the results in Table 5 and Fig. 9, The PSNR values improve appropriately with moment order increases, indicating an improvement in the reconstructed signal quality. The best quality of the reconstructed signal (lower MSE) is likewise obtained at the last moment order, as shown in Table 5 and Fig. 10.

Table 5 Performance of Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moments with Householder method at different orders
Fig. 9
figure 9

PSNR of the reconstructed signal "MIT-BIH Rec. 101" using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder methods

Fig. 10
figure 10

MSE of the reconstructed signal "MIT-BIH Rec. 101" using Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moment with Householder methods

Discussion

This paper contributes to the ongoing discussions about using DOMs in analyzing one-dimensional bio-signals. In addition, it also introduces an algorithm to overcome the propagation of numerical errors problem faces high-order computation of DOPs. The comparative experiments shown in the above tabular and graphical results assure the superiority of DOMs in reconstruction biosignals. It also demonstrates the advantages of the proposed re-orthonormalization methods (Gram–Schmidt, Householder, and Given Rotations).

Generally, the increase in polynomial order, the increase in error propagation. Therefore, many researchers used QR decomposition methods to overcome these errors. In this work, we present Discrete Orthogonal Moments (DOMs) in bio-signals analysis and reconstruction, which are gaining popularity in analyzing one-dimensional signals due to their effectiveness in capturing digital information without redundancy. The works addressed by others faced propagation errors at high order polynomials which destroy the orthogonality property of these polynomials. While in our work, the problem of error propagation at high order polynomials is solved using QR decomposition methods. Hence, the OP reconstructs the bio-signals efficiently. Moreover, to highlight the efficiency of the different forms of QR decomposition like the Gram–Schmidt method, the Householder method, and the Given Rotations method in different situations, we compare the different methods to each other to show the differences between them. Consequently, we introduce a road map to the interested researchers. Additionally, we compare five common types of moments (Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner) to estimate the best moment in analyzing and reconstructing bio-signals which gives a clear guide to researchers in this area.

The results discussion obtained can be divided into three sections. Section one is the performance of the DOMs in biosignal reconstruction and the effect of using re-orthonormalization methods in maintaining the orthogonality property at the high-order computation of DOPs. In general, the superiority of DOMs in the reconstruction of bio-signals can be attributed to the following worthwhile factors:

  • DOMs are orthogonal moments with orthogonal basis functions. Each moment coefficient can capture the signal's distinct and unique components with no information redundancy.

  • According to the order value, orthogonal moments' basis functions can extract various distinct types of information from the signals.

  • Moments generated from discrete orthogonal polynomials are effective at compressing signals. This is because they have a higher efficiency of energy compression for common signals. If the discrete orthogonal moment is chosen correctly, the energy in the signal is concentrated on a small fraction of the moment coefficients; these coefficients are then stored and used to generate the reconstructed signal.

  • The ability of DOMs on local and global feature extraction.

  • Using recursive formulas to compute polynomial values by using lower polynomial orders instead of directly computing them causes computational efficiency in the computation of the moments.

Section two determines which of the three re-orthogonalization methods (Gram–Schmidt, Householder, and Given Rotations) best preserves the orthogonality property. The comparative results indicate that the Householder method is the best in signal reconstruction in terms of reconstruction errors (PSNR, MSE) and execution time. The most likely explanation of the result has explained the fact that using Gram–Schmidt after computation of each nth Polynomial order minimizes the numerical error propagation considerably. Therefore, the Gram–Schmidt method is not stable when used in a re-orthogonalization matrix with large size. To this end, the Householder method outperforms the Gram–Schmidt and Givens rotation methods in numerical stability in the \(QR\) decomposition of a matrix with large size. In addition to Householder method is faster compared to the Gram–Schmidt and Givens rotation methods. Because of this, the Householder method is better for real-time applications. The last section investigates which type of discrete orthogonal moments (Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner) provides better-reconstructed signals. Besides that, tracking the reconstructed signal quality for DOMs at various orders of moments. The obtained results demonstrate that all of the used moment types are stable since they enabled the reconstruction of the signals until the high moment order. It reflects the effectiveness and numerical stability of the orthogonal moment for large-size signal reconstruction. This numerical stability is ensured by re-orthogonalization methods (Gram–Schmidt, Householder, and Given Rotations), especially the Householder method.

Despite the development of several reconstruction methods, substantial limitations must still be addressed. In the design of reconstruction methods, computational complexity and memory management play a crucial role, particularly in real-time applications like Remote Monitoring Systems. Reconstruction techniques increase the complexity of memory management. When the memory required to conduct the compression technique exceeds the available device memory, efficient reconstruction cannot be accomplished. Even though some reconstruction techniques achieve higher reconstruction quality, they do not manage memory effectively. Consequently, memory management and computational complexity in reconstruction techniques are interesting future research directions.

Conclusion

This article presents a method for bio-signal reconstruction based on Discrete Orthogonal Moments (DOMs). It also proposes a modified version of DOPs using the QR decomposition methods such as the Gram–Schmidt, Householder, and Given Rotations methods. The purpose of the proposed modification is to preserve the orthogonality property in the computation of high polynomials order. Based on the results, it can be concluded that the research into DOPs has been very successful. DOMs of various types: Tchebichef, Krawtchouk, Charlier, Hahn, and Meixner moments provide good results in reconstruction quality (PSNR, MSE). The comparative experiments demonstrate the superiority of the proposed modification of DOMs in reconstruction quality. This improvement in DOMs performance is due to QR decomposition methods to preserve the orthogonality property and then overcome the propagation of numerical errors. We also conclude that Tchebichef, Krawtchouk, and Charlier moments are better than Hahn and Meixner moments in reconstruction quality, and generally, Tchebichef has the best performance in signal reconstruction. The experiments of performance DOMs in reconstruction quality at a high order of moments are performed. We have noticed that the reconstruction quality improvement (PSNR highest, MSE lower) with moment orders increases. It means that the DOMs used in the proposed modification are efficient in large-size signal reconstruction. We could use the proposed method for large-size signal compression and classification in our future work and research direction. In addition, other applications will be used instead of bio-signal, such as volumetric medical images, Galaxies images, and Retrieval systems for Biomedical Images. The proposed method's ability in reconstruction could be improved by using a new version of DOMs as fractional DOMs, Radial DOMs.

Availability of data and materials

The data that used this study belong to Physionet database; fPCG signals from the fPCG Dataset are used.

Abbreviations

DOMs:

Discrete orthogonal moments

MSE:

Mean square error

PSNR:

Peak signal to noise ratio

DOPs:

Discrete orthogonal polynomials

MGSM:

Gram–Schmidt method

HM:

Householder method

GRM:

Given Rotations method

References

  1. A. Daoui et al., New algorithm for large-sized 2D and 3D image reconstruction using higher-order Hahn moments. Circuits Syst. Signal Process. 39(9), 4552–4577 (2020)

    Article  Google Scholar 

  2. G. Hassan et al., An efficient retrieval system for biomedical images based on Radial Associated Laguerre Moments. IEEE Access 8, 175669–175687 (2020)

    Article  Google Scholar 

  3. S.M.M. Rahman, T. Howlader, D. Hatzinakos, On the selection of 2D Krawtchouk moments for face recognition. Pattern Recognit. 54, 83–93 (2016)

    Article  Google Scholar 

  4. R. Benouini et al., Fast and accurate computation of Racah moment invariants for image classification. Pattern Recognit. 91, 100–110 (2019)

    Article  Google Scholar 

  5. M. Abd Elaziz, K.M. Hosny, I.M. Selim, Galaxies image classification using artificial bee colony based on orthogonal Gegenbauer moments. Soft. Comput. 23(19), 9573–9583 (2019)

    Article  Google Scholar 

  6. M. Yamni et al., Fractional Charlier moments for image reconstruction and image watermarking. Signal Process. 171, 107509 (2020)

    Article  Google Scholar 

  7. B. Xiao et al., Fractional discrete Tchebyshev moments and their applications in image encryption and watermarking. Inf. Sci. 516, 545–559 (2020)

    Article  MathSciNet  MATH  Google Scholar 

  8. F. Ernawan, N. Kabir, K.Z. Zamli, An efficient image compression technique using Tchebichef bit allocation. Optik 148, 106–119 (2017)

    Article  Google Scholar 

  9. K.M. Hosny, A.M. Khalid, E.R. Mohamed, Efficient compression of volumetric medical images using Legendre moments and differential evolution. Soft. Comput. 24(1), 409–427 (2020)

    Article  Google Scholar 

  10. M. Yamni et al., Novel Octonion moments for color stereo image analysis. Digit. Signal Process. 108, 102878 (2021)

    Article  Google Scholar 

  11. H. Zhu et al., General form for obtaining discrete orthogonal moments. IET Image Process. 4(5), 335–352 (2010)

    Article  MathSciNet  Google Scholar 

  12. M. Yamni et al., Robust audio watermarking scheme based on fractional Charlier moment transform and dual tree complex wavelet transform. Expert Syst. Appl. 203, 117325 (2022)

    Article  Google Scholar 

  13. M. Yamni et al., Efficient watermarking algorithm for digital audio/speech signal. Digit. Signal Process. 120, 103251 (2022)

    Article  Google Scholar 

  14. K.M. Hosny, A.M. Khalid, E.R. Mohamed, Efficient compression of bio-signals by using Tchebichef moments and Artificial Bee Colony. Biocybern. Biomed. Eng. 38(2), 385–398 (2018)

    Article  Google Scholar 

  15. I.S. Fathi, M.A. Ahmed, M.A. Makhlouf, E.A. Osman, Compression techniques of biomedical signals in remote healthcare monitoring systems: a comparative study. Int. J. Hybrid Inf. Technol. 1(1), 33–50 (2021)

    Google Scholar 

  16. S.H. Abdulhussain et al., On computational aspects of Tchebichef polynomials for higher polynomial order. IEEE Access 5, 2470–2478 (2017)

    Article  Google Scholar 

  17. C. Camacho-bello, J.S. Rivera-lopez, Some computational aspects of Tchebichef moments for higher orders. Pattern Recognit. Lett. 112, 332–339 (2018)

    Article  Google Scholar 

  18. E.J. Huertas, A. Lastra, A. Soria-lorente, watermarking applications of Krawtchouk–Sobolev type orthogonal moments. Electronics 11(3), 500 (2022)

    Article  Google Scholar 

  19. S.H. Abdulhussain et al., Fast recursive computation of Krawtchouk polynomials. J. Math. Imaging Vis. 60(3), 285–303 (2018)

    Article  MathSciNet  MATH  Google Scholar 

  20. I.S. Fathi et al., An energy-efficient compression algorithm of ECG signals in remote healthcare monitoring systems. IEEE Access 10, 39129–39144 (2022)

    Article  Google Scholar 

  21. G. Hassan et al., New set of invariant quaternion Krawtchouk moments for color image representation and recognition. Int. J. Image. Graph 22(04), 2250037 (2021)

    Article  Google Scholar 

  22. H. Karmouni et al., Fast and stable computation of the Charlier moments and their inverses using digital filters and image block representation. Circuits Syst. Signal Process. 37(9), 4015–4033 (2018)

    Article  MathSciNet  Google Scholar 

  23. A. Daoui et al., Stable computation of higher order Charlier moments for signal and image reconstruction. Inf. Sci. 521, 251–276 (2020)

    Article  MATH  Google Scholar 

  24. M. Yamni et al., Robust zero-watermarking scheme based on novel quaternion radial fractional Charlier moments. Multimedia Tools Appl. 80(14), 21679–21708 (2021)

    Article  Google Scholar 

  25. J.S. Rivera-lópez et al., Color image reconstruction by discrete orthogonal moment. J. Data Anal. Inf. Process. 5, 156–166 (2017)

    Google Scholar 

  26. A. Daoui et al., On computational aspects of high-order dual Hahn moments. Pattern Recognit. 127, 108596 (2022)

    Article  Google Scholar 

  27. M. Sayyouri, A. Hmimid, H. Qjidaa, A fast computation of novel set of Meixner invariant moments for image analysis. Circuits Syst. Signal Process. 34(3), 875–900 (2015)

    Article  MATH  Google Scholar 

  28. T. Jahid et al., Image analysis by Meixner moments and a digital filter. Multimedia Tools Appl. 77(15), 19811–19831 (2018)

    Article  Google Scholar 

  29. W. Ford, Numerical Linear Algebra with Applications: Using MATLAB (Academic Press, Cambridge, 2014)

    MATH  Google Scholar 

  30. A. Daoui, M. Sayyouri, H. Qjidaa, Efficient computation of high-order Meixner moments for large-size signals and images analysis. Multimedia Tools Appl. 80(2), 1641–1670 (2021)

    Article  Google Scholar 

  31. L.N. Trefethen, D. Bau III., Numerical Linear Algebra (Siam, Philadelphia, 1977)

    MATH  Google Scholar 

  32. https://www.physionet.org/physiobank/database/mitdb/.

Download references

Acknowledgements

Not applicable.

Funding

Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Author information

Authors and Affiliations

Authors

Contributions

I.S. conceived of the problem, gathered data, conducted the analysis, and produced the first draft of the paper. M.A. reorganized the paper and provided the related work. M.M. evaluated the arguments critically, reviewed the versions, and provided helpful comments. All authors provided general feedback and helped to revisions. All authors read and approved the final manuscript.

Authors’ Information

Islam S. Fathi received the B.Sc. and M.Sc. degrees in math and computer sciences from the Faculty of Science, Zagazig University, Egypt, in 2013 and 2019. He is currently working as a Teaching Assistant with the Department of Information Systems, Al Alson Academy, Cairo, Egypt. His research interests include signal processing, meta-heuristic optimization, and bioinformatics.

Mohamed Ali Ahmed is currently an assistant professor with the Mathematics Department in the Faculty of Science Suez Canal University. His research interests include Computer Networks, Performance Evaluation, Queuing Systems, meta-heuristic optimization and bioinformatics.

Mohamed Abd Allah Makhlouf is currently an associate professor in the Faculty of Computer Science and informatics at Suez Canal University and Faculty of Computer Science, Nahda University. He received his first degree in Computer Science and Operation Research, Faculty of Science, Master’s degree in Expert Systems, Faculty of Science Cairo University. He received his Ph.D. degree in computer science from the Faculty of Science, Zagazig University. He got Post-Doctoral studies in Computer Science from Granada University, Spain, in 2016. His research interests include Machine learning, data mining, Intelligent Bioinformatics, meta-heuristic optimization, Decision support systems and predictive models, Bioinformatics, meta-heuristic optimization, Decision support systems and predictive models.

Corresponding author

Correspondence to Islam S. Fathi.

Ethics declarations

Ethics approval and consent to participate

We declare that the studies in this work do not involve any human participants, human data, human tissue and animal.

Consent for publication

We declare that the manuscript does not contain any individual person’s data in any form (including individual details, images or videos).

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Fathi, I.S., Ahmed, M.A. & Makhlouf, M.A. An efficient computation of discrete orthogonal moments for bio-signals reconstruction. EURASIP J. Adv. Signal Process. 2022, 104 (2022). https://doi.org/10.1186/s13634-022-00938-4

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s13634-022-00938-4

Keywords

  • Discrete orthogonal moments
  • Bio-signal reconstruction
  • Orthogonalization methods
  • Propagation error
  • Discrete orthogonal polynomials