J. Vis. Commun. Image R.

92 (2023) 103798

Contents lists available at ScienceDirect

J. Vis. Commun. Image R.

journal homepage: www.elsevier.com/locate/jvci

Full length article

Reversible data hiding based on automatic contrast enhancement using

histogram expansion✩
WanLi Lyu a , YaJie Yue a , Zhaoxia Yin b,c ,∗
a
Anhui Provincial Key Laboratory of Multimodal Cognitive Computation, School of Computer Science and Technology, Anhui University, 230601, China
b
Shanghai Key Laboratory of Multidimensional Information Processing, East China Normal University, Shanghai, 200241, China
c
School of Communication and Electronic Engineering, East China Normal University, Shanghai, 200241, China

ARTICLE INFO ABSTRACT

Keywords: A lot of Reversible Data Hiding (RDH) methods aim to generate a stego image infinitely approaches the
Reversible data hiding original image while the quality of the original image is leaved out of consideration. Juxtaposed with a
Histogram shifting plain image, a contrast enhanced version always improves the user experience significantly. Reversible Data
Automatic image enhancement
Hiding with Contrast Enhancement (RDHCE) enhances the stego image contrast combined with its payloads
Reversible contrast enhancement
and enables the cover image to be regained accurately after the payloads have been extracted. This study
Histogram expansion
presents a novel RDHCE method using histogram expansion. First, a new local histogram selecting strategy
is proposed to improve the contrast of the whole image. Meanwhile, the global average brightness is used as
a reference to determine the shifting direction of the local histogram to prevent the image from being over-
enhanced. Moreover, the contrast can be improved adaptively when a reasonable number of data is embedded
at the selected embedding points. Experimental results show that, with a given payload, the proposed method
achieves better contrast and maintains good visual quality compared with state-of-the-arts.

1. Introduction Contrast Enhancement (CE) is a valuable image processing method

that can make image details more visible. A medical or a satellite image
Using data hiding technology to communicate in the Internet is usually has a low visual quality, and it has to show high details in
an attractive area of research [1–3]. Image data hiding methods can low light conditions. Contrast enhancement is a useful technique in
be subdivided into irreversible and reversible forms [4–6]. Irreversible this case. Histogram equalization is a common approach to enhancing
data hiding methods cause the cover image changed irrevocably after the image contrast [12,14–16], and it is often used in medical image
embedding data, and make it impossible to restore the original cover
analysis, satellite image analysis, and scientific visualization.
image. Conversely, Reversible Data Hiding (RDH) can imperceptibly
Histogram shifting technology affects histogram equalization nat-
embed additional data into image and then restore the image error-
urally. A local histogram shifting [17,18] is performed on the local
less. At present, data hiding technique has been effectively applied to
so many fields such as digital watermarking, secret communication, texture region of the image. Using the local histogram shifting method
and copyright protection. In some special cases, such as medical im- to embed data can effectively enhance the local details of the image
ages, military images, and legal forensics, the cover images should and increase the embedding capacity but easily cause visual artifacts.
be restored accurately [7–10]. In recent years, the research on RDH In contrast, a global histogram shifting can improve image quality
technology has become a very active research topic. to reduce visual artifacts. Wu et al. first proposed a reversible data
Image visual quality and embedding payload are two important hiding method that employs a histogram shifting for image contrast
indications of RDH methods. There is a mutually restrictive relationship enhancement, commonly called RDHCE [19]. When the peak bin is
between the two, when the embedding payload increases the image selected in the histogram as the data hiding position, the nearest zero
visual quality degradation appears [11]. Peak Signal-to-Noise Ratio bin is located. The pixels between the peak and zero bin shifted one unit
(PSNR) is commonly used evaluation of the stego image visual qual-
pixel towards the zero bin. The RDHCE method requires preprocessing
ity [12,13], which evaluates the image distortion after data embedding
before embedding to avoid overflow.
compared to the original cover image objectively.

✩ This paper has been recommended for acceptance by Zicheng Liu.

∗ Corresponding author.
E-mail address: [email protected] (Z. Yin).

https://doi.org/10.1016/j.jvcir.2023.103798
Received 28 November 2022; Received in revised form 19 January 2023; Accepted 26 February 2023
Available online 2 March 2023
1047-3203/© 2023 Elsevier Inc. All rights reserved.
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

More recently, a number of pretreatment methods have been pro- 2.2. BPACERDH method
posed for reversible contrast enhancement, such as maintaining or
minimizing the variation of the pixel values to reduce the distortion Kim et al. [25] first proposed a reversible data hiding method for
created by shifting [20,21], or using a threshold to reduce the dis- automatic contrast enhancement on the basis of preserving brightness.
tortion created by preprocessing [22–24]. The preprocessing stage of This method embeds secret data using more than once histogram
these approaches requires an initial value, and more importantly, the shifting, according to the mean brightness of the image. To preserve
embedding process is not automatic. brightness, the histogram shifting again according to the comparison
The automatic contrast enhancement using reversible data hid- result of the mean brightness values before and after the previous
ing (ACERDH) method was first proposed by Kim et al. [14] as an histogram shifting. Let 𝑃𝑠 be the bin containing the highest number of
automatic RDHCE. This method suffers from excessive contrast en- pixels and 𝑃𝑐 be the bin containing the lowest number closest to 𝑃𝑠 . In
hancement since it disregards the average brightness of the original the next round, the shifting direction can be decided by the comparison
image. Kim et al. [25] subsequently proposed a novel average contrast of 𝑃𝑠 and 𝑃𝑐 :
enhancement method, which uses an adaptive histogram shifting to {
preserve the original image’s brightness, and is therefore denoted as 1, 𝑃𝑠 < 𝑃𝑐 ,
𝑑= (1)
BPACERDH. Kim et al. [26] finally proposed an alternative approach to −1, 𝑃𝑠 > 𝑃𝑐 ,
enable contrast enhancement while maintaining a minimum distortion 𝑑 here is computed by comparing the pixel values of 𝑃𝑠 and 𝑃𝑐 , and
between the original and the embedded data. the result of 𝑑 shows the direction of shifting: 1 indicates the histogram
Gao et al. [27] proposed a RDH method to improve the embedding shifting to the right, while −1 indicates left.
capacity and visual quality of medical images. The region of interest Based on the shifting directions, the embedding process can be
(ROI) of the medical image is labeled at first [28], and then the divided into two types: Right histogram shifting using
histogram of the labeled ROI is stretched and shifted. After the peak
bin in the ROI histogram is selected, the nearest zero bin to the selected ⎧ 𝑝 +𝑏 , 𝑝𝑖 = 𝑃𝑠 ,
⎪ 𝑖 𝑘
peak bin is located. The pixels between the peak bin and the zero bin all 𝑝′𝑖 = ⎨ 𝑝𝑖 + 1, 𝑃𝑠 < 𝑝𝑖 < 𝑃𝑐 , (2)
shifted one unit in the direction of the zero bin, and then the data was ⎪ 𝑝𝑖 , 𝑒𝑙𝑠𝑒,
⎩
embedded into the peak bin. Subsequently, Gao et al. [29] proposed an
automatic contrast enhancement method using reversible data hiding, and left histogram shifting using
which performs a histogram expansion based on two-sided histogram ⎧ 𝑝 −𝑏 , 𝑝𝑖 = 𝑃𝑠 ,
shifting and brightness preservation. ⎪ 𝑖 𝑘
𝑝′𝑖 = ⎨ 𝑝𝑖 − 1, 𝑃𝑠 < 𝑝𝑖 < 𝑃𝑐 , (3)
Mansouri et al. [30] proposes a two-sided histogram expansion ⎪ 𝑝𝑖 ,
⎩ 𝑒𝑙𝑠𝑒,
method to enhance contrast, accordingly known as TSACERDH au-
tomatically. Their method selects two peaks in the original image will 𝑏𝑘 denotes the 𝑘th bit value (0 or 1) to be hidden. For each pixel
histogram, and at the left and right of the two peaks selects a nearest 𝑝𝑖 , the stego pixel 𝑝′𝑖 is obtained.
valley for each. All interval bins shifted one unit from the peak towards For the generation of the location map, we use 𝑀𝑜 for recording
the matching valley. The bins between the peaks and outside the valleys {
do not change, and the secret data is embedded at the two vacated 1, 𝑝𝑖 = 𝑃𝑐 − 𝑑,
𝑀𝑜 = (4)
positions. 0, 𝑝𝑖 = 𝑃𝑐 .
Inspired by ACERDH [14] and BPACERDH [25], a reversible data In the location map 𝑀 = (𝑀1 , … , 𝑀𝑜 , …), if the pixel value 𝑝𝑖 equals
hiding method based on histogram automatic contrast enhancement 𝑃𝑐 , the 𝑀𝑜 is marked with 0, and if the pixel value 𝑝𝑖 equals 𝑃𝑐 − 𝑑, the
is proposed in this study. The proposed method hides the secret data 𝑀𝑜 is marked with 1. Before each repeated shifting of the histogram,
by automatically one-way histogram shifting, enhances image contrast the location map 𝑀 is generated. After the last histogram shifting, the
combined with payloads, and does not need a preprocessing step. First, values of 𝑃𝑠 and 𝑃𝑐 are recorded in the least significant bits (LSBs) of the
the mean brightness value of the whole image is calculated and the first 16 pixels of the original image, while the LSBs of the 16 original
highest peak bin is located. And then a range in the global histogram is image pixels are added to the side information in the last histogram
set to select the local highest peak bin and the local lowest bin. At the shifting round.
same time, whether the payload capacity can hold the appropriate side
information should be determined to ensure that the original image can 3. Proposed method
be recovered accurately.
ACERDH [14] automatically selects embedding positions to embed
2. Related works
data. Still, after embedding data, the change of mean brightness is
left out of consideration, then the over-enhanced stego image will be
In this section, we briefly review the ACERDH [14] and BPAC-
obtained after the histogram shifting. To enhance the image contrast
ERDH [25] methods to explain our reversible data hiding with the
contrast enhancement method. while maintaining a steady mean brightness, BPACERDH [25] sets the
histogram shifting direction by comparing the cover mean brightness
2.1. ACERDH method with the revised mean brightness after embedding data.
We propose a novel method for automatic contrast enhancement
ACERDH [14] was the first automatic RDHCE method. ACERDH and preserving the image brightness. The main contributions of the
adopts multiple one-way histogram expansion and generates location proposed method are as follows:
maps at the same time. Each histogram shifting creates a location map (1) No parameters are to be prepared in advance, and the proposed
to recover the original image accurately. The pixel values with the method adaptively selects histogram shifting intervals to prevent the
highest and lowest bin in the global histogram are selected. The pixel stego image over-enhanced and reduce the entropy caused by the global
values with the highest bin are served to embed secret messages. The histogram shifting. After data hiding, the image enhances contrast and
pixels between the highest and the lowest shifted one unit towards the maintains a steady mean brightness.
lowest bin. When move to the end, the next to lowest bin combined (2) The bins are selected adaptively at each iteration, with the
with the lowest bin, in this way, the distortion of image modification highest peak bin shifting direction to the lowest bin, the data are
can be reduced. Repeat this procedure without manually setting a embedded, and histogram equalization is also achieved.
parameter until there is no longer enough embedding space to carry The proposed method consists of the following four phases. The first
the side information, the contrast enhancement is done. step is to locate the highest and lowest bins inside the given range, as

2
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Fig. 1. Procedure of the proposed method.

discussed in Section 3.1. The second stage creates the side information to be embedded, and the payload for each embedding is the value of
and location map, as shown in Section 3.2. The third stage, embedding 𝑃ℎ .
the payload through histogram expansion, is addressed in Section 3.3. Different from 𝑃𝑠 and 𝑃𝑐 obtained by Eq. (1) in the global histogram,
Finally, in Section 3.4, we cover the production of the side information. we find the appropriate interval in the histogram first and obtain
The fourth step is to repeat the previous three stages until the stop the values of 𝑃ℎ and 𝑃𝑙 . Then, the histogram shifting direction is
condition is reached, as demonstrated in Section 3.5. Section 3.6 shows determined by 𝑑 ′ as
{
the extraction procedure and the recovery of the image. We propose 1, 𝑃ℎ < 𝑃𝑙 ,
a method for selecting where to embed data in the adaptive range in 𝑑′ = (6)
−1, 𝑃ℎ > 𝑃𝑙 ,
order to achieve reversible data hiding of contrast enhancement. The
entire flow chart for our method is shown in Fig. 1. 3.2. Location map

3.1. Histogram shifting For each histogram shifting, we need to record side information for
accurate recovery and prepare for the following data embedding. We
At this stage, we select the highest peak (𝑃ℎ ) and the lowest (𝑃𝑙 ) specify 𝐿 to record the location map information in each shifting as
in the image histogram expansion. The motivation for using 𝑃ℎ and follows
{
𝑃𝑙 is to reduce the preprocessing steps and automate the embedding 1, 𝑝𝑗 = 𝑃𝑙 − 𝑑 ′ ,
𝐿𝑝 = (7)
process. The histogram shifting direction is according to the comparison 0, 𝑝𝑗 = 𝑃𝑙 .
results of the 𝑃ℎ and 𝑃𝑙 . We embed information in 𝑃ℎ . The pixels values For each pixel 𝑝𝑗 , our proposed method can determine the number
between 𝑃𝑙 and 𝑃ℎ are shifting one unit towards the 𝑃𝑙 direction. of original pixel values of 𝑃𝑙 and 𝑃𝑙 − 𝑑 ′ . Before each repeated shifting
⎧ of the histogram, the location map 𝐿 = (𝐿1 , . . . , 𝐿𝑝 , . . . ) is generated,
⎪ 𝑝𝑗 − 1, 𝑃𝑙 < 𝑝𝑗 < 𝑃ℎ , where the location map records the pixel values number of 𝑃𝑙 and
⎪ 𝑝 𝑗 + 𝑏𝑘 ⋅ 𝑑 ′ , 𝑝𝑗 = 𝑃ℎ , 𝑃𝑙 − 𝑑.
𝑝′𝑗 = ⎨ (5)
⎪ 𝑝𝑗 + 1, 𝑃ℎ < 𝑝𝑗 < 𝑃𝑙 ,
⎪ 𝑝𝑗 , 𝑒𝑙𝑠𝑒, 3.3. Adaptive bin selection
⎩
where 𝑝′𝑗 is the watermarked pixel and 𝑝𝑗 is the pixel before embedding Unlike previous studies, our proposed method first uses the global
data under the shifting round. Let 𝑏𝑘 ∈ {𝟶, 𝟷} be the secret message bit mean brightness value as a reference for the histogram dynamic

3
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Fig. 2. (a) Select the value of 𝑃ℎ under 𝑃𝑚 > 𝐵 condition. (b) Select the value of 𝑃ℎ under 𝑃𝑚 ≤ 𝐵 condition.

expansion selection interval. The global mean brightness value can give (2) For 𝑃𝑙 settings, we select the pixel value that occurs least
an average pixel value for the current image and also show the pixel frequently in the range.
value concentration degree. It is defined as shown in Eq. (8):
𝑃𝑙 = arg min 𝐻𝑛′ , (13)
1∑
𝐼 𝑛∈𝐽𝑃
𝑙
𝐵= 𝑝, (8) {
𝐼 𝑗=1 𝑗 [2, 𝑃𝑚 + 𝛼], 𝑃𝑚 > 𝐵,
𝐽𝑃𝑙 = (14)
where 𝐼 is the sum of all pixel values of the image. [𝑃𝑚 − 𝛽, 253], 𝑃𝑚 ≤ 𝐵.
On the other hand, the pixel value 𝑃𝑚 , which has a peak value in the It can be shown that our method employs a distinct range depending
global histogram, serves as another reference for the dynamic selection on the magnitude of 𝑃𝑚 and 𝐵. As shown in Fig. 3(a), if 𝑃𝑚 is higher
interval. Our proposed method uses 𝑃𝑚 and the global mean brightness
than 𝐵, 𝑃𝑙 is selected in the range [2, 𝑃𝑚 + 𝛼]. Fig. 3(b) shows that 𝑃𝑙 is
𝐵 to determine the embedding interval and within which right bins
selected within the range [𝑃𝑚 − 𝛽, 253] based on 𝑃𝑚 being less than 𝐵.
are selected. The histogram shifting interval and the choice of 𝑃ℎ and
At the same time, there may be numerous possible choices for 𝑃𝑙 ,
𝑃𝑙 values are described below.
(1) When set 𝑃ℎ , we select the maximum value in the given interval. we choose the pixel value closest to 𝑃ℎ .

𝑃ℎ = arg max 𝐻𝑛′ , (9)

𝑛∈𝐽𝑃 3.4. Side information
ℎ

∑
𝐼
𝐻𝑛′ = ∣∣𝑝′𝑗 = 𝑛∣∣, (10) To recover the original images, some side information has to be
𝑗=1
{ provided before each histogram shifting. In order to record merged
‖ ′ ‖ 1, 𝑝𝑗 = 𝑛, bins in a joint histogram, we need to record the information in location
‖𝑝𝑗 = 𝑛‖ = (11)
‖ ‖ 0, 𝑝𝑗 ≠ 𝑛, map 𝐿. In addition, the 𝑃𝑙−1 and 𝑃ℎ−1 values for the preceding round of
here 𝐻𝑛′ is the number of pixel values of 𝑛 equal to 𝑝′𝑗 , and the range data hiding also need to be precisely recorded. When the data is first
of 𝑛 is [0, 255]. embedded, eight zeros in binary form are used as the initial values for
{ 𝑃𝑙 and 𝑃ℎ , and also provide a stop condition for restoring the cover
[𝑃𝑚 + 1, 𝐵 + 𝛼], 𝑃𝑚 > 𝐵,
𝐽𝑃 ℎ = (12) image. Moreover, for the final embedding iteration, the values of 𝑃ℎ
[𝐵 − 𝛽, 𝑃𝑚 − 1], 𝑃𝑚 ≤ 𝐵.
and 𝑃𝑙 cannot be recorded on 𝑆 for the next iteration, because there
According to Eq. (12), the sizes of 𝐵 and 𝑃𝑚 are compared within
is no next phase; instead, they can be recorded in the image pixels of
the given range to determine the dynamic selection interval.
the first 16 pixels, and the LSBs of the first 16 pixels should also be
If the 𝑃𝑚 is greater than 𝐵, the maximum pixel value is calculated
recorded. The final side information is represented as
within the range [𝑃𝑚 + 1, 𝐵 + 𝛼], where 𝛼 is represented by 𝛿(𝐵 − 𝑃𝑚 )
and 𝛿 is a constant parameter. This is shown in Fig. 2(a). 𝑆 = 𝐿∣∣𝑃ℎ ∣∣𝑃𝑙 ∣∣16𝐿𝑆𝐵𝑠, (15)
If 𝑃𝑚 is less than or equal to 𝐵, the maximum pixel value is
calculated within the range [𝐵 − 𝛽, 𝑃𝑚 − 1], where 𝛽 is represented by where 𝐿 needs and 𝑃𝑙′ 𝑃𝑙′
− 𝑑 bits, and 𝑃ℎ−1 𝑃𝑙−1
each need 8 bits, and
𝛿(𝑃𝑚 − 𝐵). This is shown in Fig. 2(b). 16 LSBs needs 16 bits (used only for the last histogram embedding).

4
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Fig. 3. (a) Select the value of 𝑃𝑙 under 𝑃𝑚 > 𝐵 condition. (b) Select the value of 𝑃𝑙 under 𝑃𝑚 ≤ 𝐵 condition.

3.5. Stop condition

Image contrast is enhanced by embedding data in the histogram

many times. When side information is greater than or equal to the
maximum theoretical embedding capacity, or satisfies the needs of
image contrast enhancement, it is a stop condition. This is because
the reversible embedding cannot be ensured if the 𝑃ℎ cannot fulfill the
produced side information in the embedding. Then the equation to stop
the procedure is:
∑
𝐼
∣∣𝑝𝑗 = 𝑃𝑙 ∣∣ < ∣𝑆𝑚𝑎𝑥 ∣, (16)
𝑗=17

where 𝑝𝑗 is the number of pixels at a time equal to 𝑃𝑙 and 𝑆𝑚𝑎𝑥 is the

value of the maximum side information.

3.6. Extraction and recovery

In the process of restoring the cover image, the LSBs values of the
first 16 pixels are extracted according to 𝑆 to obtain the 𝑃ℎ value and 𝑃𝑙
value when the data was last embedded, according to these two values, Fig. 4. Choose the relation between different intervals of distance 𝐵 and standard
deviation according to 𝛿.
the 𝑃ℎ and 𝑃𝑙 values of the last embedding can be found, as well as the
secret data of the 𝑃ℎ embedded this time, to recover the previous step
better.
The procedure for extracting is completed step by step until a total We utilize a dataset from the Kodak lossless true color image suite [31],
of 16 zero-length sequences have been extracted. In this step, the which is to transform them into grayscale images and test them for
original LSBs values of the initial 16 pixels are substituted in the data contrast and image quality.
recovered in the previous round, and the original images are finally In Section 4.1, we use 24 gray images converted from Kodak lossless
obtained. color image kits as experiment subjects. Shifting and embedding data at
various ranges in the global histogram expansion influence the standard
4. Experiment results and discussion deviations (STD) of the images themselves. In Section 4.2, 620 images
from BOSSBase are selected for our conducting experiments at different
The following is a comparison of four approaches, and the choice embedding capacities. The proposed method analyzes the perceptual
is ACERDH [14], RDHCE [19], BPACERDH [25], and TSACERDH [30]. quality of the proposed scheme in terms of subjective and objective
The experiment uses equally distributed binary sequences of zeroes and parameters, various parameter-free biases, and comprehensive evalua-
ones to embed secret data. We estimate the load by subtracting the bit tions. In Section 4.3, we show the computational difficulty and number
number of side information from the whole number of embedded bits. of repetitions while processing images.

5
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Fig. 5. Quantitative comparison between the existing and proposed methods.

4.1. Interval selection 4.2. Imperceptibility analysis

According to the different interval ranges set in the image his- Below the proposed method uses the following indicators to assess
togram, the impact degree on the image is different. Below, we use the the performance of the improved images. For reference image quality
set of images converted to grayscale as a test demonstration. evaluation, we employ peak signal-to-noise ratio (PSNR), and structural
Our proposed method uses the mean brightness 𝐵 as a reference, similarity (SSIM) [32]. PSNR is a means of expressing the difference
with which the histogram can be divided into two parts: [0, 𝐵] and between the pixel values of the stego and the cover images so that
[𝐵, 255]. Let 𝛿 be in the range of [0, 1], and the adjustment step of 𝛿 is the differences in pixel values in images may be assessed. The visual
0.1. We use the range [𝐵 − (𝐵 − 0)𝛿, 𝐵 + (255 − 𝐵)𝛿] in the histogram to quality of the image is less than optimal, with a lower PSNR. SSIM is an
test the STD of the image. Fig. 4 shows the standard deviation gradually indicator that quantifies the similarity between two images. From the
increases when 𝛿 grows to 0.7, which indicates that different ranges standpoint of image composition, it defines structural information as
selected in the histogram expansion affect the STD of the image. In the independent of brightness and contrast and takes into consideration the
following imperceptible experiments, the proposed approach sets the structural information between neighboring pixels [33]. The value of
parameter 𝛿 as 0.1 to get a minor STD. SSIM fluctuates between 0 and 1, and the more similar a pair of images

6
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Fig. 6. Subjective analysis of the proposed method.

are compared, it is to 1. Relative contrast error (RCE) is obtained by improved picture is closer to the natural image. The value of BRISQUE
comparing the standard deviation of stego and cover images. When its is between 0 and 100, and the lower the number, the stronger the
value is greater than 0.5, the image contrast is improved; when the similarity to the natural image.
value is less than 0.5, the image contrast is diminished. The relative Table 1 displays the average values of 620 BOSSBase [36] images
entropy error (REE) is the change in the entropy value of the original under varied payloads. According to the results, as the payload grows
images and the improved images. If the REE value is greater than gradually, the values of PSNR and RCE increase, while SSIM and RMBE
0.5, the image data is increased, and if the REE value is below 0.5, fall substantially. When the payload rises, it can be observed that the
the image deteriorates. Relative Average Brightness Error (RMBE) is drop is relatively minimal, and the overall outcome is better than the
to compares the average brightness of the stego image and the cover existing methods.
image. Its value fluctuates between 0 and 1; when it approaches 1, Fig. 5(a) to Fig. 5(f) shows the index curve affected by the pure
the difference between the two is smaller [34]. For reference-free hiding rate. We can see that the proposed method can better increase
image quality evaluation, such as a reference-free spatial domain image the image contrast. When the payload increases, the PSNR and RMBE
quality assessment method (BRISQUE) [35], it can evaluate whether the of the proposed method are approaching that of the TSACERDH and

7
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

Table 1
Evaluation results (average) obtained from 620 BOSSbase images.
Algorithm PSNR(dB) SSIM BRISQUE REE RCE RMBE Payload
Proposed 39.443 0.9951 28.9727 0.5056 0.5067 0.9942
TSACERDH 39.445 0.9944 28.568 0.50786 0.5072 0.9973
BPACERDH 41.1251 0.9953 28.9537 0.5057 0.5066 0.9989 0.1
ACERDH 39.0151 0.9952 28.9727 0.50786 0.5067 0.9942
RDHCE 37.0451 0.9917 27.3446 0.5063 0.5087 0.9966
Proposed 33.977 0.9864 26.097 0.5106 0.5133 0.9897
TSACERDH 34.27 0.9848 25.3758 0.5108 0.5121 0.9944
BPACERDH 35.43 0.9866 26.0926 0.5107 0.5133 0.9989 0.3
ACERDH 32.88 0.986 26.1137 0.5108 0.5117 0.9857
RDHCE 32.56 0.9701 25.4819 0.5097 0.512 0.9934
Proposed 29.5931 0.9697 24.81 0.5173 0.5239 0.9854
TSACERDH 30.28 0.9681 24.7239 0.5168 0.5231 0.992
BPACERDH 31.1635 0.9718 24.8034 0.5167 0.5227 0.9989 0.5
ACERDH 28.643 0.9709 24.8141 0.517 0.5231 0.9762
RDHCE 28.413 0.9383 24.8438 0.5136 0.5196 0.9891
Proposed 26.5184 0.9461 23.383 0.5242 0.5367 0.9829
TSACERDH 26.7743 0.9401 23.3811 0.5239 0.5365 0.9891
BPACERDH 27.5143 0.9472 23.4029 0.5239 0.5359 0.9989 0.7
ACERDH 24.9643 0.948 23.4639 0.5244 0.5372 0.9651
RDHCE 24.4643 0.8855 23.525 0.51782 0.5278 0.9819
Fig. 7. The relationship between different loads and corresponding average number of
Proposed 23.5009 0.9044 22.2 0.5327 0.5555 0.9802 iterations.
TSACERDH 24.18 0.8967 22.1455 0.5312 0.5512 0.987
BPACERDH 24.54 0.9018 22.1976 0.5312 0.5517 0.9988 0.9
ACERDH 22.04 0.9075 22.1813 0.5324 0.5541 0.9528
RDHCE 20.79 0.7945 23.6937 0.519 0.5336 0.9681 method using histogram expansion automatically chooses an appropri-
ate local histogram, shifting the local histogram to embed the data in
the peak and enhancing the image contrast over multiple iterations.
The experimental results demonstrated that our method enhances the
BPACERDH methods. The SSIM of the proposed method is slightly
image’s contrast and improves the image quality.
lower than that of ACERDH and greater than the others. When the pay-
In the future, we will continue to explore the following directions.
load is gradually increased, our method offers a certain improvement
In our proposed method, choosing an appropriate local histogram
in brightness, showing that the method stabilizes the mean brightness.
for iteration is the critical issue. Therefore, more efficient iteration
In a comparison of RCE, the proposed method shows a significant
methods can be investigated. And in recent years, reversible data
performance as the embedded payload increases.
hiding method based on automatic contrast enhancement has attracted
We use subjective visual analysis for the proposed method, and the considerable attention. Therefore, the automatic RDHCE method’s per-
data set is three images from Kodak, Fig. 6(a) to Fig. 6(r) offers a formance should also be investigated.
more intuitive visual assessment. Visually, we can see the artifact of
the doorknob in Fig. 6(b), while a clear doorknob in Fig. 6(f). It can be Declaration of competing interest
seen in Fig. 6(i) that the average brightness of the petals shows a sharp
contrast with the surrounding texture. In Fig. 6(r), the brightness of the One or more of the authors of this paper have disclosed potential or
right forehead of the little girl has a certain improvement compared pertinent conflicts of interest, which may include receipt of payment,
with the other methods and does not excessively raise the surrounding either direct or indirect, institutional support, or association with an
brightness. entity in the biomedical field which may be perceived to have poten-
tial conflict of interest with this work. For full disclosure statements
4.3. Computational complexity and the number of repetitions refer to https://doi.org/10.1016/j.jvcir.2023.103798.Yajie Yue reports
financial support was provided by National Natural Science Foundation
In this section, we conduct experiments about running time and of China.
average number of loop repetitions under different loads with the
proposed method on the 620 images used in Table 1. When analyzing Data availability
the time complexity of the proposed method, we consider that the local
interval calculation is used in the histogram operation to decrease the The data that has been used is confidential.
running time, but the average brightness value of each cycle is com-
puted, which increases the operation time. It takes less time to run than Acknowledgments
BPACERDH and ACERDH but more than TSACERDH and RHDACE.
The relationship between different loads and the corresponding average This work was supported in part by the Natural Science Foundation
number of iterations is shown in Fig. 7. When the embedded payload of China (Grant 62172001, Grant 61872003).
is increased, the average number of watermarks increases. It is because
the data is embedded in several local histograms rather than only one References
maximum peak in a global histogram. It also can be seen from Fig. 7.
[1] H.-T. Wu, J. Huang, Y.-Q. Shi, A reversible data hiding method with contrast
However, these methods impact the iteration numbers close to the
enhancement for medical images, J. Vis. Commun. Image Represent. 31 (2015)
proposed method except for TSACERDH. 146–153.
[2] H. Yao, H. Wei, T. Qiao, C. Qin, JPEG quantization step estimation with
5. Conclusion coefficient histogram and spectrum analyses, J. Vis. Commun. Image Represent.
69 (2020) 102795.
[3] Y. Qiu, H. He, Z. Qian, S. Li, X. Zhang, Lossless data hiding in JPEG bitstream
In this study, we proposed a new RDHCE method to enhance the using alternative embedding, J. Vis. Commun. Image Represent. 52 (2018)
contrast of images. Compared to the traditional RDHCE, the proposed 86–91.

8
W. Lyu et al. Journal of Visual Communication and Image Representation 92 (2023) 103798

[4] Z. Ni, Y.-Q. Shi, N. Ansari, W. Su, Reversible data hiding, IEEE Trans. Circuits [21] H.-T. Wu, S. Tang, J. Huang, Y.-Q. Shi, A novel reversible data hiding method
Syst. Video Technol. 16 (3) (2006) 354–362. with image contrast enhancement, Signal Process., Image Commun. 62 (2018)
[5] Reversible data hiding in encrypted color images using cross-channel 64–73.
correlations, J. Vis. Commun. Image Represent. 78 (2021) 103166. [22] G. Gao, Y.-Q. Shi, Reversible data hiding using controlled contrast enhancement
[6] Z. You, S. Li, Z. Qian, X. Zhang, Reversible privacy-preserving recognition, in: and integer wavelet transform, IEEE Signal Process. Lett. 22 (11) (2015)
2021 IEEE International Conference on Multimedia and Expo, ICME, IEEE, 2021, 2078–2082.
pp. 1–6. [23] H. Chen, W. Hong, J. Ni, J. Yuan, A tunable bound of the embedding level for
[7] W. Zhang, X. Hu, X. Li, N. Yu, Recursive histogram modification: establishing reversible data hiding with contrast enhancement, in: International Conference
equivalency between reversible data hiding and lossless data compression, IEEE on Cloud Computing and Security, Springer, 2016, pp. 134–144.
Trans. Image Process. 22 (7) (2013) 2775–2785. [24] Y. Ri, J. Dong, W. Wang, T. Tan, High capacity reversible data hiding with
[8] R. Kumar, K.-H. Jung, Robust reversible data hiding scheme based on two-layer contrast enhancement, in: CCF Chinese Conference on Computer Vision, Springer,
embedding strategy, Inform. Sci. 512 (2020) 96–107. 2017, pp. 284–294.
[9] Z. Yin, L. Ke, Robust adaptive steganography based on dither modulation and [25] S. Kim, R. Lussi, X. Qu, F. Huang, H.J. Kim, Reversible data hiding with
modification with re-compression, IEEE Trans. Signal Inf. Process. over Netw. 7 automatic brightness preserving contrast enhancement, IEEE Trans. Circuits Syst.
(2021) 336–345. Video Technol. 29 (8) (2018) 2271–2284.
[10] Y. Qiu, Z. Qian, H. Zeng, X. Lin, X. Zhang, Reversible data hiding in encrypted [26] S. Kim, H.J. Kim, Automatic contrast enhancement using reversible data hiding
images using adaptive reversible integer transformation, Signal Process. 167 with distortion minimization, in: 2019 IEEE 4th International Conference on
(2020) 107288. Image, Vision and Computing, ICIVC, IEEE, 2019, pp. 291–295.
[11] Y. Yang, W. Zhang, D. Liang, N. Yu, Reversible data hiding in medical images [27] G. Gao, S. Tong, Z. Xia, B. Wu, L. Xu, Z. Zhao, Reversible data hiding with
with enhanced contrast in texture area, Digit. Signal Process. 52 (2016) 13–24. automatic contrast enhancement for medical images, Signal Process. 178 (2021)
[12] H.-T. Wu, W. Mai, S. Meng, Y.-M. Cheung, S. Tang, Reversible data hiding with 107817.
image contrast enhancement based on two-dimensional histogram modification, [28] G. Gao, X. Wan, S. Yao, Z. Cui, C. Zhou, X. Sun, Reversible data hiding with
IEEE Access 7 (2019) 83332–83342. contrast enhancement and tamper localization for medical images, Inform. Sci.
[13] H. Chen, J. Yuan, W. Hong, J. Ni, T.-S. Chen, On performance improvement of 385 (2017) 250–265.
reversible data hiding with contrast enhancement, Comput. J. 63 (10) (2020) [29] G. Gao, L. Amoah, Automatic contrast enhancement with reversible data hiding
1584–1596. using bi-histogram shifting, J. Inf. Secur. Appl. 68 (2022) 103223.
[14] S. Kim, R. Lussi, X. Qu, H.J. Kim, Automatic contrast enhancement using [30] S. Mansouri, H.K. Bizaki, M. Fakhredanesh, Reversible data hiding with au-
reversible data hiding, in: 2015 IEEE International Workshop on Information tomatic contrast enhancement using two-sided histogram expansion, J. Vis.
Forensics and Security, WIFS, IEEE, 2015, pp. 1–5. Commun. Image Represent. 81 (2021) 103359.
[15] Improving preprocessing in reversible data hiding based on contrast [31] R. Franzen, Kodak lossless true color image suite, 4, (2) 1999, Source: http:
enhancement, J. King Saud Univ.- Comput. Inf. Sci. (2021). //r0k.us/graphics/kodak.
[16] Q. Ying, Z. Qian, X. Zhang, D. Ye, Reversible data hiding with image [32] Z. Wang, A.C. Bovik, H.R. Sheikh, E.P. Simoncelli, Image quality assessment:
enhancement using histogram shifting, IEEE Access 7 (2019) 46506–46521. from error visibility to structural similarity, IEEE Trans. Image Process. 13 (4)
[17] E. Fragoso-Navarro, M. Cedillo-Hernandez, F. Garcia-Ugalde, R. Morelos- (2004) 600–612.
Zaragoza, Reversible data hiding with a new local contrast enhancement [33] Reversible data hiding in medical images with enhanced contrast in texture area,
approach, Mathematics 10 (5) (2022) 841. Digit. Signal Process. 52 (2016) 13–24.
[18] T. Zhang, T. Hou, S. Weng, F. Zou, H. Zhang, C.-C. Chang, Adaptive reversible [34] M.-Z. Gao, Z.-G. Wu, L. Wang, Comprehensive evaluation for HE based
data hiding with contrast enhancement based on multi-histogram modification, contrast enhancement techniques, in: Advances in Intelligent Systems and
IEEE Trans. Circuits Syst. Video Technol. (2022) 1. Applications-Volume 2, Springer, 2013, pp. 331–338.
[19] H.-T. Wu, J.-L. Dugelay, Y.-Q. Shi, Reversible image data hiding with contrast [35] A. Mittal, A.K. Moorthy, A.C. Bovik, No-reference image quality assessment in
enhancement, IEEE Signal Process. Lett. 22 (1) (2014) 81–85. the spatial domain, IEEE Trans. Image Process. 21 (12) (2012) 4695–4708.
[20] H.-T. Wu, Q. Huang, Y.-m. Cheung, L. Xu, S. Tang, Reversible contrast enhance- [36] P. Bas, T. Filler, T. Pevnỳ, ‘‘Break our steganographic system’’: the ins and outs
ment for medical images with background segmentation, IET Image Process. 14 of organizing BOSS, in: International Workshop on Information Hiding, Springer,
(2) (2020) 327–336. 2011, pp. 59–70.