Performance Evaluation of Modified Color Image Steganography Using Discrete Wavelet Transform

1 Introduction

The development in the Internet and multimedia processing technologies has made the distribution of digital data more easy and even at low cost. The data can be well edited with almost trifling loss using multimedia processing techniques. Therefore, the protection of sensitive data is a major issue. To resolve this problem, a number of techniques for protecting the sensitive data have been proposed. Among them, digital steganography has drawn significant attention from the research community.

Digital steganography refers to the process of hiding secret data into carrier data such that the existence of the secret data is undetectable. The digital images, audios, videos, and other files can be used as a carrier to embed the secret data. The use of digital image is of particular interest for the research community as it requires low bandwidth for exchange [5].

The image steganography techniques are broadly classified into two main categories: spatial domain and frequency domain techniques [1]. The former embeds the secret data into the least significant pixels of cover data. The least significant bit (LSB) is the most widely used spatial domain technique. The other well-known techniques are pixel value differencing (PVD) and histogram shifting. The main pros of these techniques are ease in understanding and its implementation. However, these are sensitive toward image-processing attacks.

The frequency domain steganography technique transforms the cover data into frequency domain coefficients and then embeds the secret message in it. The transformation techniques used are discrete Fourier transform (DFT), discrete wavelet transform (DWT), and discrete cosine transform (DCT). DFT-based techniques introduce round-off errors that make them unsuitable for steganography. However, a few techniques are available in literature: for example, McKeon [16] used 2D DFT for steganography. The main drawbacks of the DCT-based techniques are small embedding capacity and artifact problems. In contrast to DFT, DWT-based techniques provide both spatial as well as temporal information. They have large embedding capacity compared to DCT. They are more robust against noise and signal-processing attacks. Hence, DWT has been widely used for image steganography. A large number of DWT-based color image steganography techniques have been reported in literature [1, 5]. However, these techniques are least robust against image-processing attacks. To solve these problems, there is a need to find the optimum block replacement approach for hiding the data onto wavelet coefficients of cover color image that are robust against attacks.

This paper intends to use for color images, the same work already proposed for grayscale images, by the authors [15]. The work uses two new concepts, namely, blocking and secret key computation. The main purpose of introducing a blocking concept is to reduce the effect of an embedded secret image into the cover color image. Both cover and secret images are divided into small non-overlapping blocks. The blocks of secret images are embedded into blocks of three planes of cover color images using the best matching criteria. The secret key computation procedure has been proposed to store the address of the best-matched blocks of planes of cover image with those of the secret image based on the least error criterion. The performance of the proposed color image steganography has been tested on a variety of images and compared with several other recently proposed techniques.

The remaining structure of the paper is as follows: Section 2 presents the related works. The proposed DWT-based image steganography is presented in Section 3. The experimental results are reported in Section 4. The concluding remarks are depicted in Section 5.

2 Related Works

Numerous color image steganography techniques are reported in literature. The well-known steganography technique is the LSB substitution, which is a well-known steganography technique in which the LSB of the pixels is modified to embed the secret data [4]. The drawback of this technique is that it has a much smaller peak signal to noise ratio (PSNR) value, and the extracted secret image visual is not good.

Wu and Tsai [24] presented an edge-based steganography method that utilizes the concept of pixel value differencing. They also discovered that edge-area pixels can carry more data. This method was further extended by Chen et al. [6]. They utilized the hybrid edge detection mechanism that combines fuzzy and canny edge detector. This method provides better image quality. This method was further improved by Ioannidou et al. [10]. Ioannidou et al. [10] used Sobel and Laplacian filters instead of canny. The main drawback of this technique is the overhead of two different files that contain the information about data embedding.

Kumar and Kumar [13] proposed a steganography technique based on the combination of the DCT and the DWT. They applied the DCT on a secret image to find the DCT coefficients. Thereafter, they applied the DWT on the cover image and the DCT coefficient image to find the image features. The extracted image features of the secret image are embedded into the cover image. The main drawback of this technique is to hide the image features only in one portion of the cover image. Kumar and Kumar [14] proposed a new color image steganography technique using the DWT. They divided the cover image based on the red, green and blue (RGB) method, thereafter, hiding the secret image into the RGB of the cover image. However, the embedding capacity of this approach is low.

Karim et al. [12] proposed a novel approach based on the LSB using a secret key. The secret key encrypts the hidden information, and then, it is stored into a different position of the LSB of the image. This method provides better security than the other existing techniques. Shejul and Kulkarni [22] proposed an algorithm in which binary images are embedded inside the cover image by taking the color planes of the cover image. The secret image is inserted into the cover image by cropping the cover image according to the skin tone detection and then applying the DWT. However, the embedding capacity of this approach is too low.

Rubab and Younus [21] proposed a method using the DWT and the Blowfish encryption technique to conceal the text message in the color image. The drawback of this technique is that it is computationally expensive. Bassil [3] proposed a color image steganography that utilizes the canny edge detector. The LSBs of every edge pixel identified by the canny edge detector are replaced with secret bits. However, it did not provide the guarantee of correct retrieval of the secret data. Hemalatha et al. [9] designed an image steganography technique to hide both image and key in the color cover image using the DWT and the integer wavelet transform (IWT). However, the security of the hidden data is less. Ghebleh and Kanso [7] proposed a chaotic algorithm technique based on a three-dimensional chaotic cat map and the DWT. They embedded the cat map of a secret image into the cover image. Kanan and Nazeri [11] designed a lossless spatial domain technique that uses a genetic algorithm to find the best positions in the cover image for data embedding. However, it suffers from large computational complexity problem.

Ou and Sun [18] proposed a method for image steganography based on the absolute moment block truncation coding (AMBTC). In this scheme, a threshold computed to classify the blocks of the AMBTC compressed codes as smooth or complex blocks in which data can be embedded. The two quantization levels in smooth block are re-calculated to minimize the distortion in the image. Gunjal and Mali [8] proposed a technique that utilized both the DWT and the singular value decomposition (SVD). They applied the DWT on the cover image to decompose in different bands. Thereafter, the SVD was applied on the approximation band to further decompose. The secret data were embedded in the decomposed band to increase the security. Pan et al. [19] proposed a technique that combined compressive sensing with subsampling. The cover image tends to be compressed in the transform domain. The characteristics of compressive sensing, dimensional reduction, and random projection were utilized to insert the secret message into the compressive sensing transform domain of the sparse image. The bit correction rate between the original secret image and the extracted message was used to compute the accuracy [14, 18].

Yang and Wang [25] developed a color image steganography technique based on a smart pixel-adjustment process. The block of two adjacent color pixels is used to embed the secret data. However, the embedding capacity of this technique is not high. Baby et al. [2] proposed a data-hiding technique that utilizes the DWT for embedding the multiple color images into a single image. The secret images were embedded into three color planes of the cover images. Swain [23] proposed an adaptive PVD steganography technique. In this technique, the secret data is embedded in the block of each color plane. During the embedding process, the vertical and horizontal edges are exploited in each block. This technique works on the color plane instead of the color pixels.

Prasad and Pal [20] presented a color image steganography technique that utilizes the concept of the PVD. This technique has readjusted the overlapping blocks of the color components. The main advantages of this technique is its simplicity and is easy to implement. Muhammad et al. [17] proposed a secure image steganography technique based on the stego key-directed adaptive least significant bit (SKA-LSB) method. The stego key and secret data are encrypted using a two-level encryption and multi-level encryption algorithm, respectively. The encrypted information is embedded in the cover image using an adaptive LSB technique.

As we have seen in the study of previous works, it has been found that color image steganography algorithms are good at embedding secret messages either text or image, but there is a possibility to enhance the capacity of these techniques so that the same can carry large secret information.

3 Color Image Steganography

3.2 Modified Color Image Steganography Technique

The main contribution of this paper is to evaluate the performance of the modified steganography approach for the color images. The DWT is used to decompose both the cover and secret images. The technique consists of two procedures: embedding and extraction. The details of these procedures are illustrated as follows.

3.2.1 Embedding Procedure

Figure 1 shows the block diagram of the proposed embedding procedure. The color cover image is decomposed into three color planes (i.e. R, G, and B). The three color planes are further decomposed into four coefficients: approximation coefficients, horizontal detail coefficients, vertical detail coefficients, and diagonal detail coefficients using DWT. The secret image is also decomposed into four coefficients such as the approximation coefficients, horizontal detail coefficients, vertical detail coefficients, and diagonal detail coefficients using DWT. These coefficients are partitioned into non-overlapping blocks. The approximation coefficient blocks of the secret image are matched with the detail coefficients of the color planes of the cover image. The replacement of the secret blocks is done with the best matched detail coefficient block of the cover image. The inverse DWT is applied on the approximation and modified detail coefficients of the color image to obtain the stego image (ST).

Figure 1:

Proposed Embedding Procedure.

The steps of the embedding procedure are described in detail as follows:

1. Read the cover image (C) and secret image (S). Decompose C into three planes R, G, and B.

   C→{RC, GC, BC}

2. Apply DWT on the RGB components of C and the secret image S.

   DWT(RC)→{RCA, RCH, RCV, RCD}

   DWT(GC)→{GCA, GCH, GCV, GCD}

   DWT(BC)→{BCA, BCH, BCV, BCD}

   DWT(S)→{SA, SH, SV, SD}

3. Apply the blocking process on the approximation coefficients of the R, G, and B planes of C and the approximation coefficients of S. The approximation coefficients are decomposed into blocks of 4×4 pixels.

   SA→{BSAi, 1≤i≤BSAN}

   RCA→{BRCAi, 1≤i≤BRCAN}

   GCA→{BGCAi, 1≤i≤BGCAN}

   BCA→{BBCAi, 1≤i≤BBCAN}

4. Apply the blocking process on the detail coefficients of the R plane of C. All of these coefficients are decomposed into blocks of 4×4 pixels.

   RCH→{BRCHi, 1≤i≤BRCHN}

   RCV→{BRCVi, 1≤i≤BRCVN}

   RCD→{BRCDi, 1≤i≤BRCDN}

5. Apply the blocking process on the detail coefficients of the G plane of C. All of these coefficients are decomposed into blocks of 4×4 pixels.

   GCH→{BGCHi, 1≤i≤BGCHN}

   GCV→{BGCVi, 1≤i≤BGCVN}

   GCD→{BGCDi, 1≤i≤BGCDN}

6. Apply the blocking process on the detail coefficients of the B plane of C. All of these coefficients are decomposed into blocks of 4×4 pixels.

   BCH→{BBCHi, 1≤i≤BBCHN}

   BCV→{BBCVi, 1≤i≤BBCVN}

   BCD→{BBCDi, 1≤i≤BBCDN}

7. Each block of the secret image (BSAi) is matched with the detail coefficients of the R, G, and B planes of C. The block matching is done with the root mean square error. The position of the minimum error blocks of the R, G, and B planes are stored. The position of the best matched blocks is stored in Key K1.

8. Replace the best matched detailed coefficients of the R, G, and B planes of C with the approximation coefficients of the secret image.

9. Apply the IDWT on the modified detailed coefficients of the R, G, and B planes of C. The stego image (ST) is produced.

3.2.2 Extraction Procedure

The block diagram of the extraction procedure is shown in Figure 2. The ST is separated into three color planes. The three color planes are further decomposed into four coefficients using the DWT. These coefficients are partitioned into non-overlapping blocks. The best matched blocks are extracted from the detailed coefficients. The secret image is generated from the extracted blocks. The extraction procedure is described in detail as follows:

1. Decompose the ST into the three color planes, R, G, and B.

   ST→{RST, GST, BST}

2. Apply the DWT on all the three color planes R, G, and B planes of the ST.

   DWT(RST)→{RSTA, RSTH, RSTV, RSTD}

   DWT(GST)→{GSTA, GSTH, GSTV, GSTD}

   DWT(BST)→{BSTA, BSTH, BSTV, BSTD}

3. Apply the blocking process on the detail coefficients of the R, G, and B planes of the ST. All of these coefficients are decomposed into blocks of 4×4 pixels.

4. The position of the embedded blocks of the secret image is found using the Key K1. These embedded blocks are the approximation coefficients of S.

5. Apply the IDWT on the extracted approximation blocks and detailed coefficients of S. This will produce the extracted secret image from the ST.

Figure 2:

Proposed Extraction Procedure.

4 Performance Evaluation

The performance of the proposed technique is evaluated on the different cover and secret images. The experimental results are assessed and compared with four recently developed steganography techniques.

5 Conclusions

A modification in the DWT-based color image steganography approach has been used. The approach uses blocking and secret key computation concepts. The blocking concept uses the least variation concept. The secret key uses the concept of the detail coefficient of the DWT and the least error matching criteria. The experimental results indicate that the modified approach provides better ST and secret images in terms of PSNR. It helps us to provide better visual quality. Moreover, this approach does not require the original cover image to extract the secret image.

Figure 3:

Original Cover Images: (A) Goldhill, (B) Lena, (C) Barbara, (D) Plane.

Figure 4:

Original Secret Images: (A) Cameraman, (B) Baboon.

Figure 5:

STs After Embedding the Cameraman Image: (A) Goldhill, (B) Lena, (C) Barbara, (D) Plane.

Figure 6:

STs After Embedding the Baboon Image: (A) Goldhill, (B) Lena, (C) Barbara, (D) Plane.