91 Original scientific paper  MIDEM Society Journal of Microelectronics, Electronic Components and Materials Vol. 47, No. 2(2017), 91 – 99 Reversible Data Hiding Based on Radon and Integer Lifting Wavelet Transform A.Amsaveni1, P.T Vanathi2 1Kumaraguru College of Technology, Department of Electronics and Communication Engineering, Coimbatore, India 2PSG College of Technology, Department of Electronics and Communication Engineering, Coimbatore, India Abstract: This paper presents a reversible data hiding technique based on radon and integer lifting wavelet transform to secure the data transmitted over communication network. The technique focuses on three optimality criteria, namely imperceptibility, robustness, and reversibility. The frequency domain strategy is applied due to its superior performance over the spatial domain techniques in certain important aspects like robustness and reversibility towards signal processing and image processing operations. The cover image is first transformed from spatial domain to radon domain and then, this radon image is applied with integer lifting wavelet transform. As the Radon transform performs rotation, scaling and translation operations on the cover image, it changes the locations of the secret bits. Hence, it is very difficult to detect the embedded data without taking the inverse Radon transform and subsequently, it increases the security of the embedded payload. The integer lifting wavelet transform guarantees complete reversibility as they produce integer wavelet coefficients. Then, the middle bit planes of high frequency lifting coefficients are compressed using arithmetic coding to provide space for embedding secret payload. As the proposed framework embeds data in red, green, and blue channels, it can work well for a variety of images with different distribution of colors. Keywords: reversible data hiding; radon transform; integer lifting transform;bit plane coding;arithmetic coding Reverzibilno skrivanje podatkov na osnovi Radonove in diskretne valčne transformacije Izvleček: Članek opisuje tehnike reverznega skrivanaj podatkov na osnovi Radonove in diskretne valčne transformacije za varovanje podatkov preko omrežja. Tehnika sloni na trhe kriterijih: neopaznost, robustnost in reverzibilnost. Uporabljena je strategija na osnovi frekvence saj prednjači pred prostorsko tehniko v ključnih točkah robustnosti in reverzibilnosti obdelave signalov in slik. Slika je najprej pretvorjena v radon proctor in nato še z diskretno valčno transformacijo. Radonova transformacija opravi rotacijo, skaliranje in translacijo in spremeni lokacijo skrivnih bitov. Brez inverzne transformacije je skoraj nemogoče odkriti skrite podatke. Valčna transformacija zagotavlja popolno reverzibilnost in določi valčnbe koeficiente, ki so stisnjeni z aritmetičnim kodiranjem. Predlagan postopek vsebije morer, zelen in rdeč kanal, tako da je uporaben za številne slike. Ključne besede: reverzibilno skrivanej podatkov; radonova transformacija; diskretna valčna transformacija; koditanje bitne ravnine; aritmetično kodiranje * Corresponding Author’s e-mail: amsaveni.a.ece@kct.ac.in 1 Introduction Securing data transmitted over the internet has be- come a challenging issue caused by the advancement in data digitization and communication networking over the past decade. Therefore, it is necessary to de- vise strategies to secure information during the  pro- cess of information exchange. Reversible data hiding has emerged as a major research area due to the phe- nomenal growth in internet and multimedia technolo- gies. It involves concealing confidential data within an- other seemingly innocuous cover media such as text, video, audio, images, and compression coding [1]. 92 A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 Basic terminologies used in data hiding are as follows: Secret Payload – Message to be embedded in the cover image; Cover image – Image that carries the secret mes- sage; Stego image / Embedded image – Cover image after embedding the secret payload (Cover image + Secret data); Imperceptibility – Measure of distortion which is caused by embedding the secret message in the original cover image. It is the inability of the human eye to differentiate between the cover image and the embedded image. Generally, embedded images with higher imperceptibility are preferred in data hiding; Robustness – Robustness is the ability of the secret pay- load to withstand various intentional attacks such as image processing operations and unintentional attacks such as addition of noise; Embedding Capacity – Data embedding rate or number of bits embedded per pixel, measured in terms of bits per pixel; Security / Undetect- ability – Any data hiding system may be considered as secure if the possibility of knowing the presence of a secret message in any cover medium is very difficult; Attacks – Process of revealing the hidden data from the embedded image by attacking with various signal pro- cessing and image processing techniques. While embedding secret information in a cover image, the emphasis is on two key problems. The first one is to produce an embedded image with a tolerable level of quality so that the distortion produced due to data embedding is imperceptible. The second is to pro- duce the embedded image that is distortion tolerant (robust), i.e., even if the embedded image is attacked during communication, the hidden data can be recov- erable. Robustness of data hiding techniques can be enhanced if the properties of the cover image could be properly utilized. By considering these aspects, embed- ding data in frequency domain becomes more popu- lar compared to spatial domain techniques [2]. The frequency domain techniques modify the frequency coefficients of the cover image by applying a specific transformation function on the cover image. They are designed to be imperceptible and robust against vari- ous geometrical transformations and external attacks. Frequency domain schemes use transformation meth- ods, such as integer cosine transform [3] or integer wavelet transform [4] to compute the transform coef- ficients of the cover image. Then, these coefficient val- ues are modified to embed the secret data. In revers- ible data hiding technique based on DCT, one secret bit was embedded into two neighboring DCT coefficients in an image block [5].The secret payload is embedded in the high frequency coefficients of discrete wavelet transform (DWT) by exploiting the statistical properties of the cover image [6]. The DWT works well against var- ious image processing attacks. As it produces floating coefficients, embedded data is potentially lost while re- constructing the cover image by inverse wavelet trans- form. This drawback is overcome by integer wavelet transform (IWT). The secret data is embedded into the middle frequency of the integer lifting wavelet domain by modifying the histogram of the cover image [7]. Data embedding is also performed in radon domain, which shows a considerable improvement in bit er- ror rate. A radon-based approach was introduced to incorporate translation invariance properties to the payload [8]. RST invariant watermarking technique has been proposed by utilizing the Fourier transform and transforming them to log polar coordinates, which are quite flexible towards rotation, scaling and translation attacks [9]. Hybrid transform has been proposed based on the unique features of the transforms in the hybrid combi- nation, so that it is able to address the robustness and reversibility criteria. Accordingly, the rotation, scaling and translation properties of the Radon transform and reversibility property of integer lifting transform have been joined together in a hybrid formation. The paper investigates a combinatorial data hiding approach us- ing radon transform and integer lifting wavelet trans- form (ILWT). Radon transform ensures robustness and ILWT makes the algorithm reversible by using the lift- ing scheme on orthogonal and bi-orthogonal wave- lets. The superiority of this combination is also tested and compared with the other existing works in litera- ture. The proposed method improves the quality of embedded image as well as the robustness of embed- ded payload against various attacks compared to the existing methods. This paper is organized as follows. Section 2 describes the Radon transform. Section 3 explains Integer Lifting Transform. Section 4 discusses the proposed algorithm. Section 5 presents the experimental results and, Sec- tion 6 concludes the paper. 2 Radon transform Radon transform is a linear transform which is an effec- tive method to analyze signal between the spatial do- main and its projection space. It represents the image as a collection of projections along various directions [10]. It computes the projection of the image intensity along a radial line oriented at a specific angle. For each angle q and at each distance ρ, the intensity of a ray perpendicular to the ρ axis is summed up at R (ρ,q). Ra- don transform converts rectangular coordinates (x, y) into polar coordinates (ρ,q). The simplest form of dis- crete Radon transform is to select finite number of the 93 angular variable of projection, then to take the summa- tion on the discrete image along the projection line. As shown in Fig.1, the radon transform of a two dimen- sional function f (x, y) is the integral of function f along a straight line parallel to the y-axis, which is given by, ( )'R xθ = ( )' ' , ' ' 'f x cos y sin x sin y cos dyθ θ θ θ − − −∫ (1) Where ' ' x y       cos sin x sin cos y θ θ θ θ        −    = (2) Figure 1: Geometry of the Radon transform An efficient reversible data hiding method must be ro- bust against a wide range of image processing opera- tions such as image enhancement, cropping, rotation, scaling, compression, and signal processing operation such as addition of noise. However, conventional data hiding algorithms are more sensitive to geometric distortions. Hence, radon transform is introduced to perform rotation, scaling, and translation operations on the cover image. These operations change the posi- tioning of the secret bits. Without taking inverse Radon Transform, it is very difficult to detect the embedded data and subsequently, this increases the security of embedded payload. 3 Integer lifting wavelet transform There are many researches that have been explored us- ing wavelets in the field of image processing and image steganography. The main advantage of wavelet is that they offer multi-resolution capability, which is similar to the operation of the human visual system. Wavelets provide an optimal representation of signals. Normally, wavelet-based data hiding gives better performance compared to other methods. As the conventional wavelet transform performs a convolution of the input image and the wavelet basis, it requires large memory space for the computation process. The time taken and the large memory required for the conventional wave- let transform is reduced in the lifting method. Lifting transform is a technique used in constructing second generation wavelets entirely in the frequency domain. It is fast compared to the first generation wavelets, as it requires only addition and subtraction. Conventional wavelet transform is not suitable for re- versible data hiding scheme as reversibility property is not guaranteed. Wavelet transform operates on a floating-point arithmetic basis. An image that has inte- ger intensity values in the spatial domain is converted into decimal wavelet coefficients. The wavelet coeffi- cients are modified appropriately during data hiding operation and inverse wavelet transform is carried out to reconstruct the image in the spatial domain. A se- rious note here is that practically wavelet coefficients are truncated or rounded as it is not viable to represent the coefficients to its full accuracy. Information is po- tentially lost while reconstructing the image by inverse wavelet transform. But reversible data hiding schemes have to recover the host image without distortion along with the secret payload. Eventually, this makes the discrete wavelet transform a poor choice for revers- ible data hiding [11]. To address this specific issue, an invertible integer lifting wavelet transform is used in the proposed scheme. The system operates on integer arithmetic and alleviates the loss of any information via forward and inverse transforms [12]. The lifting wavelet transform decomposes the image into frequency subbands, which contain approxima- tion and detail coefficients. The system reserves the detail coefficients, which have texture, edges, and re- gion boundary for data hiding. It is an insensible region for human visual system. An advantage of the Lifting Scheme is that it can be converted easily into a trans- form that maps integers to integers while retaining the perfect reconstruction property. Thus, embedding data in integer lifting wavelet domain satisfies the proper- ties like security, imperceptibility, and robustness of the proposed technique. 4 Bit plane embedding using binary arithmetic coding The Binary Arithmetic Coding can be exploited for compressing the bit planes of grayscale/colour images. As arithmetic coding is a lossless compression method, it guarantees the recovery of original payload. In bit plane embedding, the most significant bits for each A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 94 pixel are grouped into one bit plane, the next most significant bits into another bit plane and so on till the least significant bit plane. Mostly, the five highest order planes contain visually significant data. The other lower bit planes contain fine details in the image. Lesser the bit plane number, lesser is its contribution to the final stage. Statistically, there is an equal distribution of zeroes and ones in the lower planes of the image than in the high- er planes. This leads to lower compression ratio and lower embedding capacity in the lower bit planes than in the higher planes. This is because binary sequences of length L having higher probability may be encoded more compactly than another one of the same length with a lower probability. But the signal to noise ratio falls down as higher bit planes are altered for embed- ding [13]. The most significant bit plane contains the most critical approximation values of the image. Hence, modifica- tions made in higher bit-plane may degrade the quality of the cover image. In order to have the embedded im- age visually as same as the cover image, data is hidden in one or more middle bit planes. The bits in one or more bit-planes can be compressed to provide space to hide data like text or image due to the existence of redundant information. The ap- proximate coefficients in the LL sub-band contribute to visual perception. Hence, the secret bits are embed- ded in LH, HL or HH subbands (Detail Coefficients). The original bits in the selected bit plane of LH, HL or HH subbands are compressed using arithmetic coding to provide space for embedding the payload bits. The structure of the embedded bit plane is shown in Fig.2. CHH, CHL, and CLH headers represent the header information of the compressed HH, HL, and LH sub bands. They describe the bit distribution required for arithmetic encoder and decoder. CHH, CHL, and CLH lengths denote the length of the compressed bit stream in the chosen bit plane of the LH, HL, and HH subbands. 5 Proposed methodology Majority of the methods discussed in the literature addressed only a few of the desired characteristics, namely lossless/reversible, imperceptible, high pay- load capacity and robustness, and not all. The pro- posed method of reversible data hiding is based on bit plane embedding in radon and integer lifting wavelet domain. It aims to meet all the desired characteristics to an optimal level. The block diagram of the proposed data embedding algorithm is shown in Fig.3. Apply Radon Transform to red/ green/ blue chanel Integer Liing Wavelet Transform Select LH, HL and HH bands Choose middle bit plane(s) Apply arithmec coding Embedding algorithm Take Inverse ILWT and Inverse radon transform Combine all the three channels Approximaon Co-efficients Modified Detail Co-efficients Stego Image Message to be embedded Compressed LH, HL&HH Cover Image Figure 3: Block diagram of embedding process CLH Header (16 bits) CHL Header (16 bits) CHH Header (16 bits) CLH Length (16bits) CHL Length (16bits) CHH Length (16 bits) Length of embed- ding data (32 bits) Secret Payload Figure 2: Structure of the embedded bit plane A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 95 5.1 Data Embedding Algorithm Input: Cover image with M-rows and N-columns, Se- cret payload bits. Output: Stego Image Step 1: Read the cover image of size M x N. Step 2: Separate color channels (Red, Green, and Blue) of the cover image. Step 3: Apply radon transform on any one of the channels. Step 4: Notice that Radon image undergoes a single level integer lifting wavelet transform which results in 4 subbands (LL, LH, HL, HH) of size M/2 x N/2, each. Step 5: Construct binary images from the chosen bit planes of LH, HL, and HH bands. Step 6: Compress the original bits in the chosen bit plane of these bands using arithmetic coding and obtain the header information required for the arithmetic encoder and decoder. Step 7: Read the secret payload and convert it into a bit string. Step 8: Concatenate the header length, header infor- mation, and compressed bit streams of CLH, CHL, and CHH and the secret payload to get a single bit stream. Step 9: Embed bit stream into chosen bit plane of LH band. If not enough, embed in HL band and then in HH band and observe that it results in embedded LH, HL, and HH components. Step 10: Apply inverse integer lifting transform on LL coefficients and the modified LH, HL, and HH coefficients. Step 11: Compute inverse radon transform of the im- age obtained from Step 10. Step 12: Combine all the three color channels to get stego image. 5.2 Data Extraction Algorithm Step 1: Read the stego image of size M x N. Step 2: Separate color channels (Red, Green, and Blue) of stego image. Step 3: Apply radon transform to the channel in which the data is embedded. Step 4: Notice that Radon image undergoes single level integer lifting wavelet transform which results in 4 subbands (LL, LH, HL, HH) of size M/2 x N/2, each. Step 5: Construct binary images from chosen bit planes of LH, HL, and HH bands. Step 6: Derive the header information and header length needed for arithmetic decoding. Step 7: Extract the compressed bits from the chosen bit plane of these bands using arithmetic de- coder and decompress the subbands to get the reconstructed subbands. Step 8: Apply inverse integer lifting transform on the reconstructed subbands LH, HL, and HH along with LL subband. Step 9: Compute inverse radon transform of the im- age obtained from Step 8. Step 10: Combine all the three channels to get the original cover image. 6 Expermental results In order to investigate the performance of the pro- posed data hiding algorithm, several experiments are carried out in a computer system equipped with Intel core 2 duo processor with 2 GB memory and a clock speed of 2 GHz. Matlab 8 (R2013a) platform is used for the digital simulation of the algorithm. Five standard 512 x 512 color images such as (a) Airplane,(b) Baboon, (c) Boat, (e) Lena and (e) Pepper, obtained from USC-SIPI (Image database), have been used as cover images. The performance of the algorithm is investigated in terms of imperceptibility, robustness, and reversibility. 6.1 Imperceptibility The metrics used to test the imperceptibility property of the proposed algorithm are PSNR (Peak Signal to Noise Ratio) and SSIM (Structural Similarity Index Meas- ure). The PSNR for an image of size M x N is calculated by, PSNR = 10 log10 (255 2 / MSE) dB (3) where, MSE= ( ) ( )( ) 1 1 , ' , M N x y p x y p x y = = −∑∑ 2 (4) where p(x, y) stands for the pixel value in the cover im- age and p’(x, y) is the pixel value at position (x, y) in the stego image after embedding the secret message. M and N denote the number of rows and columns of the image and (x, y) denotes the pixel coordinates. The quality of the stego image is also calculated using SSIM as follows: SSIM = ( ) ( ) ( )( ) 2μμy c1 2σxy c2 μ2x μ2y c1 σ2x σ2y c2+ + + + ++ (5) where x and y are same size windows of the cover and stego images and μx and μy are corresponding x and y averages. σ2x and σ 2 y are the variances of x and y and A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 96 σxy is the covariance of x and y. The positive constants c1 and c2 are included to avoid a null denominator. Typically c1 = (k1L) 2 ; c2 = (k2L) 2; L= (2no.of bits/pixel) -1; [k1, k2]=[0.01,0.03] by default. Table 1 shows the PSNR and SSIM of cover images after embedding a payload of 10,000 bits using the wavelet cdf2.4 (Cohen-Daubechies-Feauveau 2.4) in the fourth bit plane of Red, Green and Blue channels. As the PSNR obtained from the stego images is greater than 42 dB, the embedded payload is highly imperceptible to the human eye, i.e., the perceptual quality of the result- ant stego images is good. The red channel offers bet- ter PSNR and SSIM compared to green and blue chan- nels. The red channel gives an improvement in PSNR of about 2.0 to 7.0 dB over green and blue channels. Among all cover images, the Airplane image yields bet- ter PSNR for the same payload. Table 1: Quality metrics Cover image PSNR (dB) and SSIM Red Channel Green Channel Blue Channel PSNR SSIM PSNR SSIM PSNR SSIM Airplane 47.70 0.9771 47.65 0.9867 46.70 0.9519 Baboon 43.29 0.9900 42.91 0.9704 42.49 0.9894 Boats 46.96 0.9921 43.23 0.9740 43.69 0.9693 Lena 45.87 0.9820 42.46 0.9780 43.46 0.9409 Pepper 46.86 0.9784 44.34 0.9840 42.39 0.9656 The Fig.4. shows the embedded images of size 512 x 512, after embedding a payload of 10,000 bits using cdf2.4 in the fourth bitplane of red channel of standard images. Table 2 summarizes the quality of the Lena image un- der varying payload, after embedding in the bit plane 4, 5, and 6 of red channel using wavelet cdf2.4. Table 2: Embedding capacity Vs. PSNR (dB) Secret payload (bits) Embedding Rate (bpp) PSNR (dB) Bit Plane 4 Bit Plane 5 Bit Plane 6 1000 0.004 45.27 43.08 41.42 3000 0.011 45.26 43.07 41.40 6000 0.023 45.14 43.04 41.20 10,000 0.038 45.10 43.02 41.06 20,000 0.076 44.68 42.54 40.83 40,000 0.153 42.21 40.41 38.33 50,000 0.191 42.20 39.80 38.33 70,000 0.267 41.49 39.52 37.44 80,000 0.305 41.28 38.79 37.17 86,000 0.328 41.16 38.72 37.13 90,000 0.343 Insufficient 38.64 37.11 95,000 0.362 Insufficient Insufficient 37.04 96,000 0.366 Insufficient Insufficient Insufficient The perceptual quality of cover image will get reduced if the data is embedded in higher bit planes and also PSNR drops down as more number of bits embedded in that plane. The bit plane 4 can accommodate only 90,000 bits as the bit plane 4 provides less space for data embedding compared to other bit planes. The maximum embedding capacity of bit planes 5 and 6 is 95,000 bits and 96,000 bits respectively. Beyond 95,000 bits, there is no space to accommodate the secret pay- load in both the bit planes. The embedding capac- ity completely depends upon cover image and is also based on the bit distribution of the chosen bit plane. The PSNR varies from 41.12 dB at the embedding rate of 0.004 bits per pixel to 37.04 dB at 0.362 bits per pixel for bit plane 6. The experimental results of the proposed scheme are compared with the various schemes discussed in Tsai & (a) (b) (c) (d) (e) Figure 4: Standard color images of size 512 x 512 after embedding a payload of 10,000 bits (a) Airplane; (b) Ba- boon; (c) Boat; (d) Lena; (e)Pepper A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 97 Sun [14], Fu & Shen [15], and Niu et al [16] and summa- rized in Table 3. The PSNR of the cover images is meas- ured after embedding a payload of 10,000 bits in Red channel using the wavelet cdf2.4.The PSNR offered by the proposed scheme is about 15% greater than Niu et al scheme. The proposed scheme gives better PSNR for Lena image compared to Baboon and Boat images. Table 3: Comparison of the proposed scheme with the schemes in the literature Host image PSNR (dB) Tsai and Sun Scheme Fu and Shen Scheme Niu et al Scheme Proposed Scheme Lena 39.43 38.10 40.57 45.87 Baboon 41.76 38.90 41.67 43.29 Boat 42.49 39.72 40.71 44.03 6.2 Robustness Robustness is measured using Bit Error Rate (BER) and is defined as: k k 0 b b n k= = ∑BER ’ / N (6) where b and b’ are embedded and extracted bits re- spectively, N is the total number of secret bits embed- ded and represents the XOR operation. The value of BER ranges between 0 and 1. If BER is closer to 1 then, it means that the error value of extracted data is higher. The value of BER is calculated after retrieving the se- cret data from the embedded block. Lower the BER%, higher is the accuracy of the extracted secret data. The stego images are added with Gaussian noise with a variance of 0.2 and 0.4, Poisson Noise, Impulse Noise with a variance of 0.05 and 0.1 and Speckle Noise with a variance of 0.05 and 0.1. Generally, addition of noise is responsible for the degradation of the image. The im- age processing operations such as Rotation (5 and 10 degrees), Scaling (200% and 400%), Blurring and Crop- ping (10% and 25%) are performed on embedded im- ages. After subjecting to the attacks, the original cover image is extracted and the bit error rate of extracted payload over secret payload is measured. Table 4 summarizes the experimental results for the proposed data hiding scheme against various attacks. As the BER is about 0.15 to 0.35 % of embedded payload, the algorithm is robust against various intentional and unintentional attacks. 6.3 Reversibility In order to ensure the reversibility, the extracted cover image and the original cover image must be similar.The metric used to measure the similarity between the two images is Normalized Correlation Coefficient (NCC). The value of 0 represents no correlation. NCC will ap- proach to one if the extracted cover image resembles the original cover image. The Normalized Correlation Coefficient between cover image and extracted cover image is defined as, ( ) ( )( ) ( )( ) ( )( ) , 2 2 , , ( , ) , , , − − = − − ∑ ∑ ∑ mean meanx y mean meanx y x y f x y f g x y g f x y f g x y g NCC (7) Table 4: Effect of various attacks on BER Attacks Bit Error Rate Airplane Baboon Boats Lena Pepper Gaussian Noise (σ2 =0.2 ) 0.0032 0.0034 0.0035 0.0032 0.0034 Gaussian Noise (σ2 =0.4) 0.0036 0.0036 0.0037 0.0037 0.0038 Poisson Noise 0.0028 0.0030 0.0029 0.0030 0.0031 Impulse Noise (σ2=0.05) 0.0017 0.0020 0.0021 0.0018 0.0022 Impulse Noise (σ2=0.10) 0.0019 0.0022 0.0024 0.0020 0.0024 Speckle Noise (σ2=0.05) 0.0018 0.0021 0.0025 0.0026 0.0024 Speckle Noise (σ2=0.10) 0.0019 0.0024 0.0027 0.0027 0.0026 Rotation (5°) 0.0018 0.0021 0.0020 0.0022 0.0024 Rotation (10°) 0.0020 0.0023 0.0023 0.0024 0.0026 Scaling (200%) 0.0016 0.0017 0.0018 0.0019 0.0022 Scaling (400%) 0.0018 0.0019 0.0020 0.0020 0.0024 Blurring (5) 0.0024 0.0023 0.0020 0.0019 0.0023 Blurring (10) 0.0027 0.0026 0.0022 0.0021 0.0025 Cropping (10%) 0.0026 0.0028 0.0030 0.0029 0.0031 Cropping (25%) 0.0031 0.0034 0.0033 0.0031 0.0033 A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 98 where f (x, y) is the original cover image, g (x, y) is the extracted cover image. Table 5 summarizes the experimental results for the proposed data hiding scheme against various attacks. As the NCC values are greater than 0.98, it is concluded that the algorithm restores the original cover image ex- actly at the destination. 7 Conclusion In this paper, a reversible data hiding technique based on radon and integer lifting wavelet transform is pre- sented. Hybrid transform has been proposed based on the unique features of the transforms, so that it is able to address the robustness and reversibility criteria. As the Radon transform performs rotation, scaling and translation operations on the cover image, it changes the positioning of the secret bits. The integer lifting wavelet transform guarantees complete reversibility. The original bits in the selected bit planes of LH, HL or HH subbands are compressed using arithmetic cod- ing to provide space for embedding the payload bits. Generally, middle bit planes are used for embedding as they provide a balanced trade-off between embedding capacity and visual quality of that stego image. Data is embedded in red, green, and blue channels of the color image independently. As the PSNR obtained for the stego images is greater than 42 dB, the embedded payload is imperceptible to the human eye. The results have been compared with the existing works in the literature and the proposed method gives 10 to 15% improvement in PSNR. As the BER is about 0.15 – 0.35 % of embedded payload, the algorithm is robust to at- tacks. From the simulation results, it is inferred that the proposed algorithm exhibits reversibility due to high NCC values. 8 References 1. Amsaveni, A. and Vanathi, P.T. (2015) ‘A compre- hensive study on image steganography and steganalysis techniques’, Int. J. Information and Communication Technology, Vol. 7, Nos. 4/5, pp.406–424. 2. Amsaveni, A. and Vanathi, P.T. (2015) ‘An efficient reversible data hiding approach for colour imag- es based on Gaussian weighted prediction error expansion and genetic algorithm’, Int. J. Advanced Intelligence Paradigms, Vol. 7, No. 2,pp.156–171. 3. Yang, B., Schmucker, M., Funk, W., Busch, C & Sun, S (2004), ‘Integer DCT-based reversible watermark- ing for images using companding technique’, Proceedings of Electronic Imaging, Science and Technology, vol. 5306, pp. 405-415. 4. Xuan, G., Zhu, J., Chen, J., Shi, Y.Q., Ni, Z., & Su, W (2002), ‘Distortion less data hiding based on inte- ger wavelet transform’, IEEE Electronics Letters., vol. 38, no. 25, pp. 1646-1648. 5. Lin, S.D., Shie, S.C., & Guo, J.Y., (2010), ‘Improving the robustness of DCT based image watermark- ing against JPEG compression’, International Jour- nal of Computer Standards and Interfaces, vol. 32, no. 1, pp. 54-60. Table 5: Effect of various attacks on NCC Attacks Normalized Correlation Co-efficient Airplane Baboon Boats Lena Pepper Gaussian Noise (σ2 =0.2 ) 0.9817 0.9843 0.9825 0.9819 0.9824 Gaussian Noise (σ2 =0.4 ) 0.9850 0.9842 0.9860 0.9805 0.9814 Poisson Noise 0.9829 0.9844 0.9832 0.9822 0.9814 Impulse Noise (σ2=0.05) 0.9839 0.9843 0.9816 0.9811 0.9813 Impulse Noise (σ2=0.10) 0.9810 0.9816 0.9804 0.9802 0.9807 Speckle Noise (σ2=0.05) 0.9833 0.9822 0.9871 0.9864 0.9846 Speckle Noise (σ2=0.10) 0.9811 0.9804 0.9810 0.9820 0.9825 Rotation (5°) 0.9880 0.9806 0.9802 0.9824 0.9816 Rotation (10°) 0.9863 0.9794 0.9788 0.9804 0.9810 Scaling (200%) 0.9835 0.9820 0.9869 0.9846 0.9863 Scaling (400%) 0.9825 0.9806 0.9832 0.9828 0.9822 Blurring (5) 0.9854 0.9810 0.9809 0.9817 0.9826 Blurring (10) 0.9825 0.9806 0.9789 0.9738 0.9805 Cropping (10%) 0.9732 0.9726 0.9727 0.9728 0.9716 Cropping (25%) 0.9702 0.9706 0.9701 0.9706 0.9678 A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99 99 6. Xiang, S., Kim, H.J., & Huang, J (2008), ‘Invariant Image watermarking based on statistical features in the low frequency domain’, IEEE Transactions on Image Processing, vol. 14, no. 12, pp. 2140- 2150. 7. Xuan, G., Zhu, J., Chen, J., Shi, Y.Q., Ni, Z., & Su, W., (2002), ‘Distortion less data hiding based on inte- ger wavelet transform’, IEEE Electronics Letters., vol. 38, no. 25, pp. 1646-1648. 8. Stankovic, S, Djurovic, I & Pitas, I 2001,  ‘Water- marking In The Space/Spatial-Frequency Domain Using Two-Dimensional Radon-Wigner Distribu- tion’, IEEE Transactions on Image Processing, vol. 10, no. 4, pp. 650-658. 9. Lin, C, Wu, M, Bloom, J, Cox, I, Miller, M & Lui, Y 2001, ‘Rotation, scale, and translation resilient wa- termarking for images’, IEEE Transactions on Im- age Processing, vol. 10, no. 5, pp. 767-782. 10. Kim, H, Baek, Y, Lee, H & Suh, Y 2003, ‘Robust image watermark using Radon transform and bispectrum invariants’, Lecture Notes in Comput- er Science, pp. 145-159. 11. Lee, S, Yoo, CD & Kalker, T 2007, ‘Reversible image watermarking based on integer-to-integer wave- let transform’, IEEE Transactions on Information Forensics and Security, vol. 2, no. 3, pp. 321-330. 12. Calderbank, AR, Daubechies, I, Sweldens, W & Yeo, B 1998, ‘Wavelet transforms that map integers to integers’, Applied and Computational Harmonic Analysis, vol. 5, no. 3, pp. 332-369. 13. Xuan, G, Zhu, J, Chen, J, Shi, YQ, Ni, Z & Su, W 2002, ‘Distortion less data hiding based on integer wavelet transform’, IEEE Electronics Letters., vol. 38, no. 25, pp. 1646-1648. 14. Tsai, HH & Sun, DW 2007, ‘Color image watermark extraction based on support vector machines’ Journal of Information Sciences, vol. 177, no. 2, pp. 550-569. 15. Fu, YG & Shen, RM 2008, ‘Color image watermark- ing scheme based on linear discriminant analysis. Computer Standard & Interfaces’,vol. 30, no. 3, pp. 115-120. 16. Niu, PP, Wang, XY, Yang, YP & Ming, Y 2011, ‘A novel color image watermarking scheme in non- sampled contourlet-domain’, Journal of Expert Systems with Applications, vol. 38, no. 3, pp. 2081- 2098 . Arrived: 31. 03. 2017 Accepted: 12. 06. 2017 A.Amsaveni et al; Informacije Midem, Vol. 47, No. 2(2017), 91 – 99