Combining Haar Wavelet and Karhunen-Loeve Transform for Robust and Imperceptible Data Hiding Using Digital Images

1 Introduction

Nowadays, growth in technology such as computers and computer network offers widespread use of multimedia contents such as digital image, audio, and video [17, 20, 22, 23, 24, 25, 27, 29]. This growth has also made easy the duplication and distribution of these multimedia data. Therefore, protection of multimedia content has become an essential and difficult job. Cryptography, steganography, and digital watermarking are the important methods for protection of multimedia contents [27]. The important differences between these three methods are illustrated in Table 1 [27]. The image watermarking technique is a security tool for hiding digital information into multimedia cover documents. The hidden watermark can be later extracted or detected for the purpose of multimedia document security. As discussed in Table 1, the watermarking technique is better than other methods (cryptography and steganography) for protection of multimedia contents.

The key characteristics of the digital watermark are imperceptibility, robustness, security, data pay load, and computational complexity [25]. However, it is noticed that these characteristics are hindering each other. Therefore, it is clear that there are different watermarking techniques for improving one or a subset of these characteristics, but they compromise with other remaining characteristics. The popular application of digital watermarking is depicted in Figure 1 [25].

Figure 1:

Potential Applications of Digital Watermarking.

Depending on the type of multimedia data to be watermarked, the watermarking methods is defined as text, image, audio, and video watermarking [23, 25]. Out of four multimedia data types, data embedding capacity of the image is better than other media. In this context, the present work considers image as cover media. Image watermarking is divided into spatial and transform domain techniques. However, the transform domain techniques are more robust than the spatial domain techniques, as reported by various surveys [17, 20, 22, 23, 24, 27, 29]. The important transform domain techniques are discrete wavelet transform (DWT), Karhunen-Loeve transform (KLT), discrete Fourier transform, and singular value decomposition (SVD).

Recently, the higher robustness of watermark has been achieved by using wavelet based watermarking [1, 17, 20, 22, 23, 24, 25, 27, 29]. The DWT is a filter based system, which decomposes an image into a set of four non-overlapping multi-resolution sub-bands. The human eyes are much more sensitive to low-low (LL) sub-band of DWT; the watermark can be embedded into the other three sub-bands to maintain better visual quality of the image. The process can be repeated to obtain multiple scale wavelet decomposition. Further, it is observed that the energy of an image is concentrated in the high decomposition levels corresponding to the perceptually significant low frequency coefficients; the low decomposition levels accumulate a minor energy proportion, thus being vulnerable to image alterations. Therefore, watermarks requiring great robustness are embedded in higher level DWT sub-bands. However, DWT has poor directional information, shift sensitivity, and lack of phase information.

The subsequent part of the paper is structured as follows: Section 2 presents a brief literature review of some state-of-the-art techniques. Section 3 summaries the current state-of-the-art techniques. Section 4describes the proposed watermarking technique. Experimental results along with performance analysis are presented in Section 5. Section 6 provides concluding remarks of overall work followed by future scope of the work.

2 Literature Review

In this section, we are presenting a detailed review of related and recent watermarking methods.

Ali Al-Haj [1] proposed a combined robust and imperceptible image watermarking method using a combination of DWT and discrete cosine transform (DCT). The pseudo-noise (PN) sequence is generated for the gray scale watermark information before embedding into the cover image. Further, the sequence number is embedded into the DCT transform of the cover image. The visual quality and robustness performance of the method is evaluated and found to be robust for different known attacks. The authors in [14] present a robust and secure watermarking method through DWT, Arnold transform, and code division multiple access (CDMA). Initially, the Y component of the YIQ color model is decomposed by DWT, and the scrambled binary watermark information is encrypted by using the CDMA to generate a random number sequence. This sequence is now embedded into the selected sub-band of the DWT cover. The robustness of the method is evaluated for different attacks. Simultaneous embedding of robust and fragile watermark on the cover image using pseudo-random sequence based bit substitution is proposed by Shen and Chen [21]. Initially, the method decomposes the cover image up to third level DWT where the fragile and robust watermark is embedded into the significant and non-significant coefficient of the DWT, respectively. The method is extensively evaluated for various attacks. A robust and secure logo watermarking technique through SVD and fractional wavelet packet transform (FRWPT) is presented by Bhatnagar et al. [4]. Initially, the considered host image is decomposed into the different sub-bands by FRWPT, and SVD is applied on the FRWPT transformed coefficients. The singular value of the transformed FRWPT coefficients is modified with the singular value of the watermark image. The performance of the method is extensively evaluated in terms of peak signal-to-noise ratio (PSNR), NC, and computational and time complexity. Arsalan et al. [3] developed a blind medical image watermarking method using genetic algorithm (GA) and integer wavelet transform (IWT). In the block (size=16) based embedding process of the watermark, compression function is applied on IWT coefficients, and the watermark bit is embedded into the compressed coefficient. The compression function is only applied on those IWT coefficients which are greater than or equal to a chosen threshold value. Bouslimi et al. [6] proposed a combined encryption and watermarking technique for verifying the reliability of the echographic image through Rivest Cipher 4 (RC4), quantization index modulation (QIM), and spatial domain least significant bit (LSB) technique. The experimental results demonstrated that the performance of the method is evaluated in terms of entropy and PSNR. Lei et al. [16] proposed a watermarking technique for binary image through Haar wavelet. Initially, the cover binary image is divided into different blocks, whereas each block is also divided into the embedding and level area. Further, the method determined the number of black pixels for every block. The method first determines the flippable pixels through Haar wavelet and these pixels are embedding into the selected area of the cover image. Rosiyadi et al. [19] describe a robust and non-blind image watermarking method for copyright protection of e-government documents through DCT, SVD, and GA. In the embedding process of the logo watermark, DCT is applied on cover e-government document image using the space-filling curve for the DCT coefficients of the cover. Further, SVD is applied on each area of the DCT coefficients having different frequencies in a rectangular shape. The SVD coefficient of the cover document is modified by the control parameters consisting of the left singular vectors and singular values of the DCT-transformed logo watermark to avoid the false-positive problem as suffered by the SVD based watermarking techniques. In addition to that, the scaling factor is optimized by using the GA. The experimental results demonstrated that the performance of the method is extensively evaluated for all four areas of the DCT coefficients and found to be robust for the 19 different attacks. Bhatnagar et al. [5] proposed a robust DWT based watermarking method where the logo watermark of size 32×32 is embedded into the significant wavelet coefficient of the cover image using ZIG-ZIG sequence. The experimental result demonstrated that the performance of the method is evaluated in terms of PSNR, NC, and time efficiency (embedding and extraction) for different images. For the copyright, protection through image watermarking method is proposed by Lang and Sun [15] using fractional Fourier transform (FRFT) and hyperchaos. The middle coefficients of the FRFT transformed cover image are considered to embed binary watermark. The experimental results established that the method is robust for different attacks. Horng et al. [12] presents a robust and secure blind watermarking technique for the protection of e-governance document using DCT, SVD and GA. In the embedding process of the watermark logo image, DCT is applied on the gray scale cover image. All the DCT coefficients of the cover image are scanned into the four different blocks (from lower to higher) in a zigzag way. Further, SVD is applied on each block of the DCT transformed image where the singular value of each block of the DCT coefficient is modified with quantizing value using GA. The motivation behind using the GA is to improve the PSNR and NC performance of watermarking algorithms. Rahmati et al. [18] proposed watermarking method to protect verification documents for E-commerce application. In this method, a person identification number considered as watermark is embedded into the cover digital card image using block based watermarking algorithm. The performance (PSNR and average error ratio) of the method is evaluated for print scan attack. Region based multiple watermarking method is presented in [2] for securing the medical information through DWT, SVD, and LSB technique. The patient information considered as robust watermark is embedded in the region of noninterest (RONI) of the image through the transform domain techniques which provide the confidentiality and authenticity of the image. However, the logo watermark is considered as fragile watermark and is embedded into the region of interest (ROI) part of the cover image through LSB technique to provide the image integrity. Experimental results established that the performance of the method is extensively evaluated for various attacks. Dual level security for the medical applications is also provided through combined encryption and watermarking technique proposed by Kannammal and Rani [13]. In the embedding process of the medical image watermark, the watermark is embedded into the selected sub-band of the natural cover image. Further, the security of the watermarked is enhanced by using three different encryption techniques applied on the watermarked image. The performance of these three encryption techniques is compared in terms of time to encrypt and decrypt the message. The experimental results established that the method is robust for different kinds of attacks, and the RC4 encryption technique performs better than the other two encryption techniques. Singh et al. [26] proposed a robust medical image watermarking method through DWT and SVD. The method is embedding the image and text watermark simultaneously into the DWT cover image for patient recognition purposes. Further, the bit error rate (BER) performance of the method is reduced by applying four different error correcting codes (ECCs) on the text watermark of size 20 characters before embedding into the medical cover image. The performance of the method is extensively evaluated for various attacks with and without using the ECCs. The experimental results demonstrated that the performance of hybrid ECCs consisting of Bose–Chaudhuri–Hocquenghem and repetition code is better than other ECCs. Singh et al. [24] proposed a secure spread spectrum based text watermarking technique, where four different medical text watermarks are embedded simultaneously at different DWT sub-bands of the medical cover image to solve the medical data management issues. Further, the security of the medical information is enhanced by using encryption technique before the information is embedded into the cover image. The method is robust for various signal processing attacks. The proposed method can embed 104 characters without degradation of the visual quality of the watermarked image. Chen and zhao [8] developed a robust and blind watermarking technique for 3D images using contourlet transform and depth-image-based rendering (DIBR). The watermark generated through spread spectrum method and each watermark bits is embedded into the selected coefficients of the cover contourlet sub-bands through proper quantization. In [10] authors present a JND (just noticeable distortion) DCT based visible watermarking technique, where JND is used to rectify the distortion made by the embedding process. For watermark embedding, grey scale cover image is divided into nonoverlapping blocks, and DCT is applied on the different blocks. To achieve the embedding strength, the mapping is performed on the watermarking intensity range and the cover image’s intensity range using JND technique. An improved spread transform dither modulation based robust and secure watermarking technique is proposed by Cao et al. [7]. The watermark is only embedded into the selected embedding subspace. The security and robustness performance of the method is extensively evaluated for estimation of projection vector and amplitude scaling attacks, respectively. Zolotavkin and Juhola [30] proposed a robust watermarking method using QIM. The performance of the method is measured by watermark to noise ratio (WNR) and document to watermark ratio (DWR). The method is found to be robust. It provides high robust for additive white Gaussian noise and gain attack. Wang and Allebach [28] proposed a halftone image watermarking in which watermark is embedded into the halftone by using synchronization pattern. The performance of the method is evaluated in terms of PSNR, normalized human visual system (HVS) mean square error, and watermark rate and found to be good visual quality and achieved high watermark capacity. Essaidani et al. [9] present a robust image watermarking method using Delaunay triangulation and the feature points of the cover image. The sobel edge detector is used to determine the feature points which are used to produce Delaunay triangulation. A region based robust and secure watermarking method is presented by Sharma et al. [20] for medical applications. The method initially uses DWT and DCT to embed multiple watermark information into the cover medical image. Further, the security of the image and text watermark information is enhanced by message digest (MD5) hash algorithm and Rivest-Shamir-Adleman (RSA), respectively, before embedding into the medical cover image. In order to enhance the robustness of the text watermark, Hamming error correction code is also applied on the encrypted watermark. The experimental results have shown that the method is robust for important signal processing attacks.

3 Current State-of-the-Art Watermarking

The current state-of-the-art in wavelet based image watermarking as available in the literature is given below.

1. Various methods [1, 2, 4, 12, 14, 19, 24, 26] are using fusion of two or more transform domain techniques for improving the robustness of watermark(s).

2. Some methods [2, 13, 21, 26] enhanced the security of the watermark by embedding multiple watermarks with the same cover image or applied encryption techniques [6, 13, 20] on watermark/cover.

3. The methods proposed in [24, 26] use ECCs for addressing the channel noise distortion leading to faulty watermark and improving the robustness of the extracted watermark.

4. Region [ROI and non-region of interest (NROI)] based watermarking method is presented in [2, 17, 20] for securing important medical information. According to the robustness requirement, a different region is selected for embedding the watermark(s).

5. A GA based watermarking method is proposed in [3, 12, 19] for improving the robustness of the watermark and imperceptibility of the watermarked image.

6. In [8], an author has developed a robust and blind watermarking technique for 3D images.

A summary of various state-of-the-art techniques is presented in Table 2.

4 Proposed Technique

In this paper, a robust and imperceptible watermarking technique is proposed using fusion of DWT and KLT instead of applying DWT or KLT individually. The KLT for digital images has been discussed detail in [11]. Figure 2A and B show the watermark embedding and extraction process, respectively. The extraction algorithm is just the reverse process of the embedding algorithm. The cover image is decomposed by DWT, and KLT transform is applied on the selected sub-band of the cover. The watermark information transformed by DWT is embedded into the KLT coefficients of the DWT cover image. In order to enhance the visual quality of the watermarked image and reduce the bandwidth requirements, different de-noising dictionary based method is applied on the image. It is noticed that adaptive based dictionary method is better than DCT and global dictionary method. The method is also robust for different image processing attacks.

Figure 2:

Watermark (A) Embedding and (B) Extraction Process.

5 Experimental Result and Analysis

In this section, the performance of the proposed watermarking technique in terms of PSNR and NC has been investigated. For testing the visual quality (determined by PSNR) of the watermarked image and robustness (determined by NC) of the possibly distorted watermark, image MATLAB is used. The weighted peak signal to noise ratio (WPSNR) [25] is a modified version of the PSNR. Further, WPSNR uses noise visibility function to determine how much texture exists in any area of an image. In the proposed technique cover image of size 512×512 [5], the image watermark of different sizes (512×512, 256×256, and 64×64) are used for testing. Figure 3A shows the cover image and Figure 3B–D shows watermark images different sizes. The performance of the proposed technique is extensively determined in Tables 3–5 . It is noticed that larger gain factor results in stronger robustness of the extracted watermark, whereas smaller gain factor provides better visual quality of the watermarked image. It is quite apparent that the size of the watermark affects the quality of the watermarked image. However, degradation in quality of the watermarked image will not be observable if the size of watermark is small. Table 3 shows the performance of the proposed technique that is evaluated at different gains along with three different de-noising techniques. It is noticed that the PSNR performance with adaptive dictionary based de-noising method is better than the other two. However, the NC values are always greater than 0.9710 at the selected gain factors ranging from 0.04 to 0.1. Due to better robustness performance of the proposed technique, we are selecting the gain=0.1. Table 4 shows the PSNR and NC performance of the proposed technique with different sizes of watermark images. It is observed that the PSNR and NC values are always greater than 31 dB and 0.9583, respectively. Table 5 shows the NC performance of the proposed technique for different attacks. Refereeing this table, it is observed that the NC value is always greater than 0.9437.

Figure 3:

(A) Cover Image and Watermark Image of Size (B) 512×512, (C) 128×128, and (D) 64×64.

6 Conclusions and Future Directions

In this paper, we have presented a brief introduction of digital watermark, important applications, and a detailed literature review of different state-of-the-art watermarking methods. The summary of some existing techniques is presented in tabular format. In addition, a robust watermarking technique using fusion of discrete wavelet transform (DWT) and KLT for digital images is proposed. Further, visual quality of the watermarked image is enhanced by using different image de-noising techniques. The purpose of combined DWT-KLT is to improve the robustness of the watermark at acceptable visual quality of the watermarked image which is the prime objective of the research. However, it may have increased the computational complexity to some extent which needs to be investigated separately. The computational complexity of the proposed watermarking method can be minimized by selecting and exploring the other wavelet instead of DWT. In future, the performance of the method can also be improved with directional transform techniques, different error correction codes, and machine learning/GAs. In addition, the robustness performance of the method can also be determined with standard benchmark software.

We would like to further improve the performance, which will be reported in future communication.