Page 1 of 22

European Journal of Applied Sciences – Vol. 11, No. 5

Publication Date: October 25, 2023

DOI:10.14738/aivp.115.15576

Ching-Yu, Y., Tsung-Hsiang, H., & Pei-Yun, C. (2023). Combined Use of Enhanced Coefficient Adjustment Technique and Module- based Substitution to Hide Secret Message in ECG Signal. European Journal of Applied Sciences, Vol - 11(5). 141-162.

Services for Science and Education – United Kingdom

Combined Use of Enhanced Coefficient Adjustment Technique

and Module-based Substitution to Hide Secret Message in ECG

Signal

Ching-Yu, Yang

National Penghu University of Science and Technology, Taiwan

Tsung-Hsiang, Hsu

National Penghu University of Science and Technology, Taiwan

Pei-Yun, Chen

National Penghu University of Science and Technology, Taiwan

ABSTRACT

We present an effective electrocardiogram (ECG) steganography technique for

concealing patient’s diagnosis and sensitive information within ECG signals. The

method utilizes the enhanced coefficient adjustment technique (ECAT) and module- based substitution to embed a large number of secret bits into an ECG host. The

proposed method consists of two stages. In stage I, ECAT attempts to embed (n − 1)

bits in a host bundle with the size of 1  (n − 1) if the first decision rule is satisfied;

otherwise, the number of bits L1 = ⌊(n − 1)log2u⌋ is embedded in the bundle by Mod

u. In stage II, if the second rule is true, then the next input (n − 1) bits are hidden

into the bundle with the size of 1  n by ECAT; otherwise, data bits in length of L2 =

⌊nlog2u⌋ are embedded in the bundle by Mod u. Simulations demonstrate that the

payload and signal-to-noise ratio (SNR) of our method surpass those of existing

methods. Additionally, our method ensures good perceived quality and offers

protection against attacks. The proposed method finds application in biometric

measurements, including portable healthcare devices and IoT-based health

monitoring systems.

Keywords: Data hiding, ECG steganography, enhanced coefficient adjustment technique,

Module-based Substitution.

INTRODUCTION

With the widespread adoption of high-speed backbone networks, the convenience of 5G

wireless communication, and the ubiquitous applications of the Internet of Things (IoT),

individuals and organizations can easily share resources and conduct fast business transactions

over the Internet. However, the data transmitted over the Internet can be intercepted or

tampered with. While encryption/decryption techniques are commonly used to secure data,

they may not be suitable for devices with limited computing capabilities and storage, such as

portable or smart devices. In such cases, data hiding techniques provide an alternative way to

protect sensitive or important information due to their simplicity and low cost.

Page 2 of 22

Services for Science and Education – United Kingdom 142

European Journal of Applied Sciences (EJAS) Vol. 11, Issue 5, October-2023

Data hiding can be broadly categorized into steganography [1-3] and digital watermarking [4-

6]. Steganography focuses on providing high hiding capacity while maintaining suitable

perceived quality, as marked media with good perceived quality does not attract attention from

third parties. Typical applications of steganography include covert communications between

two parties and hiding private or important data within multimedia. However, digital

watermarking aims to provide robustness, with most watermarking schemes capable of

resisting manipulations such as noise addition, cropping, and geometric attacks. Digital

watermarking is commonly used for purposes such as copyright protection, ownership

authentication, and prevention of intentional or unintentional manipulation of multimedia.

Recent research has focused on hiding patient’s diagnosis and personal data within biometric

measurement signals such as electrocardiogram (ECG) and electroencephalography. In this

context, only related lossy ECG steganography techniques are surveyed here.

Yang and Wang [7] proposed two different approaches in ECG steganography based on the

coefficient alignment technique. Their experiments demonstrated average SNR/payload values

of 54.00 dB/7,500 bits for the high-quality version and 42.47 dB/14,783 bits for the high- capacity version. Jero et al. [8] utilized digital wavelet transform and singular value

decomposition, combined with a continuous ant colony optimization scheme, to embed

patient’s data in a two-dimensional (2D) ECG host. Simulations showed a percentage residual

difference (PRD) of 0.0018 and a peak signal-to-noise ratio (PSNR) of 62.87 dB, with a payload

size of 0.89 Kbytes. The method also demonstrated tolerance to cropping and noise addition

attacks. Yang and Wang [10] employed an absolute-value-decision rule to hide bits in an ECG

host. According to the predetermined rule, a maximum of (n − 1) data bits could be hidden in a

block of size (1 × n) using the offset coefficients. The resulting SNR/payload values for the

method were 47.54 dB/19,614 bits and 57.80 dB/9,982 bits for the 2-bit and 1-bit versions,

respectively.

Pandey et al. [11] proposed an ECG steganography method that utilizes a coupled chaotic map

with least significant bit (LSB) substitution. The method achieved a PRD of 0.21 and a PSNR of

56.83 dB with a payload size of 21 Kbytes. Additionally, by reducing the payload size to 2.4

Kbytes, the method was able to achieve a PSNR of approximately 70 dB. Yang et al. [12]

introduced a 2D approach for ECG steganography. In this method, the input ECG data is

converted into blocks of size n × n before the embedding process. Each host block can

accommodate a maximum of 2n data bits. Simulations demonstrated that the average SNR of

the method outperforms existing techniques, while the payload size remains competitive.

However, one drawback of this method is the waste of residual blocks that are incapable of

hiding bits after the conversion of ECG from 1D to 2D.

Yang and Wang [13] introduced an ECG steganography method that utilizes the coefficient

adjustment technique (CAT) within the integer wavelet transform domain to hide a secret

message. To embed data bits in the coefficients of a host block with the size of 1  n, they used

the adjusting rule, |

∑ sji

n−2

i=0

n−1

− sj(n−1)

| ≤ τ, where  is a control integer, n is the size of the host

block, and sji is the ith coefficient of the jth block. Namely, a data bit 1 or 0 can be hidden in a

host block of size 2 if −  (sj0 − sj1

)  0

or

0  ( − )  j0 j1

s s

is satisfied. Subsequently, the next

Page 3 of 22

143

Ching-Yu, Y., Tsung-Hsiang, H., & Pei-Yun, C. (2023). Combined Use of Enhanced Coefficient Adjustment Technique and Module-based Substitution

to Hide Secret Message in ECG Signal. European Journal of Applied Sciences, Vol - 11(5). 141-162.

URL: http://dx.doi.org/10.14738/aivp.115.15576

input bit 1 or 0 can be embedded in the block of size 3, if either

( ) 0 2 2

0 1

−  − 

+

j

j

s j

s

 s

or

 − 

+

0 ( ) 2 2

0 1

j

j

s j

s

s

is satisfied. Furthermore, coefficient adjustment was employed to address

violations that may occur in blocks. Theoretically, data bits of length (n − 1) can be concealed

in a host block of size n within this framework. Banerjee and Singh [14] introduced a deep

learning approach using a long short-term memory recurrent neural network (LSTM-RNN)

combined with encryption techniques for hiding secret messages in ECG hosts. The method

effectively reduces distortion between the original and predicted signals. Simulations

demonstrated that the average PSNR of the method is approximately 80 dB, surpassing that of

existing techniques. However, the payload capacity of the method is limited as data bits are only

embedded in the normal TP-segments of an ECG signal.

In a later work, Yang et al. [15] improved upon their previous method [13] by utilizing the

enhanced coefficient adjustment technique (ECAT) in ECG steganography. Instead of hiding two

data bits in a host bundle with a size of 1 × 3, the ECAT method embeds four bits in a host block

of the same size. This results in a bit-embedding efficiency of 4/3 for ECAT, which is twice as

large as the 2/3 efficiency of the CAT method proposed by Yang and Wang [13]. Moreover, the

ECAT method allows for the embedding of a large number of secret bits in an ECG host when

combined with LSB substitution. Simulations confirmed that the average payload and SNR (or

PSNR) of the ECAT method are superior to those of existing schemes.

The motivation of this study is to present an improved ECG steganography technique for the

protection of personal privacy. The advantages of the proposed method are capable of yielding

a high perceived quality and high hiding storage. In addition, our method possesses a certain

degree of robustness, which most conventional ECG steganography lack. Moreover, the

resultant SNR (or PSNR) and payload of our method are superior to those of the existing

techniques. The remainder of the paper is organized as follows: Section 2 describes the

proposed bit embedding, bit extraction, and provides a discussion and analysis. Section 3

presents demonstrations of our method, performance comparisons with existing techniques,

additional characteristics, and steganalysis of marked ECGs. Finally, Section 4 concludes the

work.

PROPOSED METHOD

To further improve the performance of the existed steganographic methods, the proposed

enhanced CAT (ECAT) equipped with the module-u substitution (Mod u) was utilized.

According to the following two predetermined rules, the proposed method consists of two

stages. In stage I, the ECAT attempts to embed (n − 1) bits in a host bundle with the size of (1 

(n − 1)) if the first following rule (R1) is satisfied; otherwise, the number of input bits L1 =

⌊(n − 1)log2u⌋ is hidden into the bundle by Mod u. The ⌊.⌋ is a floor function, and u is an integer.

In stage II, if the second rule (R2) is true, then n data bits are hidden into the bundle by the

ECAT; otherwise, data bits in length of L2 = ⌊nlog2u⌋ would be embedded in the bundle by Mod

u. Without loss of generality, let the size of a host bundle be equal to 3. Let

R1: |xi − xi+1

| < 2τ (1)

and