A Secure Invisible Watermarking Using Rijndael Algorithm and Wavelet Transform

Digital watermarking hides secret or personal information in host digital data to demonstrate and protect the copyrights of digital products, to authenticate the contents of digital data, or to convey side information such as access control or annotations. There are several fundamental requirements for watermarking such as: Perceptual invisibility. For robustness, a watermark should be resistant to a variety of manipulations, either unintentional or malicious. The detection should be accurate and especially the mean square error rate should be very small. To help protect the copyright and data security, Rijndael algorithm will be used using many mathematical operations like (Byte Substitution, ShiftRow, MixColumn and AddRound Key). The wavelet transform or wavelet analysis is probably the most recent solution to watermarking Rijndael code. Meaning by factoring technique for invisible watermarking Rijndael code is calculated and inputted in random locations. At the end, a detection process based on back propagation neural network will be used to detect watermarking string


Introduction
Information hiding represents a class of processes used to embed data into various forms of media such as images [1].The embedded data should be invisible to a human observer.The term hiding can refer to either making the information imperceptible or keeping the existence of the information secret.Depending on what 1029 information and in which form it is hidden in the media one can select robust watermarking using Rijndael ciphering algorithm.Wireless LANs have gained strong popularity in a number of vertical markets, including the health-care, retail, manufacturing, warehousing, and academia [2].These fields have profited from the productivity gains of using hand-held terminals and notebook computers to transmit real-time information to centralized hosts for processing.

Simulation Techniques
In this paper, two keys are used for this simulation: 1) Pseudo random sequence for watermarking code.
2) A Rijndael secret key (password Rijndael key).The locations, for watermarking code between sender and receiver via wireless LAN network, depend on the designer.The transmitter applies pseudorandom sequence code and Rijndael secret key (password key).The location for watermarking code is then scaled by a new technique called meaning by factoring method which depends on a mean value around watermark location.Discrete Wavelet Transform (DWT) is then repeatedly adding watermark Rijndael code to the sub bands (LH, HL and HH).This new technique gives a good invisible watermarking Rijndael code by using the above method.

Meaning By Factoring Method
In this section, a new watermark embedding and extraction system is presented, in an attempt to capture the various systems and configurations that have been presented.Many different types of watermarks have been proposed for a variety of applications, e.g., copyright image protection, broadcast monitoring, owner identification, proof of ownership, copy control and covert communication.The watermarking range lies between: where r and c are the length and width of the image , is embedded into the original image: To create a watermarked image, H' should be visually close to 1030 H.A secret key, K, may be used as shown in Fig. 1.
In watermark detection process, the watermark W' is extracted from the 'tested' image then W' is compared with the original signature W, see Fig. 1.The most common watermark embedding rules are the following: where the summation of all pixels around watermark is divided by the number of locations, called the average, as shown in Fig. 2.
The first row, first column, final row and final column should be avoided for watermarking location because there is no mean values (average values) at these locations.This new technique is called meaning by factoring, see Fig. 3.
In its simplest form, such a process has three inputs, the pseudorandom key, Rijndael password key and watermarking string as shown in the Fig. 4-.This process may be represented as blocks and this is called an encoder or embedding process.Depending on the intended application, two additional steps may be performed during the embedding and extraction process from perceptual secret analysis.These include pseudo random watermark key and Rijndael password key generation.Note that the attacker is not able to detect watermark code because of a good security achieved through design as will be shown in the results.The block diagram of the extraction process is shown in Fig. 5.

Rijndael Algorithm Specification
Rijndael algorithm is an iterated block cipher with variable block length and variable key length [4].The block length and the key length can be independently specified to 128, 192 or 256 bits with the constraint that the input and the output have the same length.Internally Rijndael operations are performed on a two dimensional array of bytes called the state.All the intermediate cipher and inverse cipher results are stored in the state.This array has four rows.The number of columns represents the data block length to be encrypted divided by 32 and is denoted by Nb.At the start of the cipher and inverse cipher operations, the input block is copied into the state array; the cipher or inverse cipher operations are then conducted on this state array.Fig. 6 shows many mathematical operations within Rijndael ciphertext algorithm.

Figure 6 Rijndael ciphertext algorithm
The cipher key is similarly considered as a rectangular array with four rows.The number of columns is equal to the key length divided by 32, and denoted by Nk.These representations are illustrated in Fig. 7 Plaintext  The number of rounds is denoted by Nr, and depends on the values of Nb and Nk as given in Table (1).For example when Nk=4, Nb=6, then Nr=12.

Table (1) Number of rounds Nr as a function of the block and key length i. The ShiftRow Transformation
In ShiftRow, the rows of the State are cyclically shifted over different offsets.Row 0 is not shifted, Row 1 is shifted over C1 bytes, row 2 over C2 bytes and row 3 over C3 bytes.The shift offsets C1, C2 and C3 depend on the block length Nb.The different values are specified in Table (2).The operation of shifting the rows of the State over the specified offsets is denoted by: ShiftRow (State).The inverse of ShiftRow is a cyclic shift of the 3 bottom rows over Nb-C1, Nb-C2 and Nb-C3 bytes respectively so that the byte at position j in row i moves to position ( j + Nb-Ci) mod Nb [6].

ii.The MixColumn Transformation
In MixColumn, the columns of the State are considered as polynomials over GF( 82 ) and multiplied modulo x 4 + 1 with a fixed polynomial c( x ), given by: a( x ) = '03' This polynomial is co-prime to x 4 + 1 and therefore invertible.This can be written as a matrix multiplication.Let s'(x ) = a(x ) ⊗  s(x ), As a result of this multiplication, the four bytes in a column are replaced by the following: ( ) The application of this operation to all columns of the State is denoted by MixColumn (State).Fig. 9 illustrates the effect of the MixColumn transformation on the State [6].This transformation is illustrated in Fig. 10.

Figure 10 In the key addition the Round
Key is bitwise exored to the State.Addroundkey is its own inverse.

Watermarking Used in the
Wavelet Domain A subband filtering system allows a signal to be separated into different frequency bands by employing a combination of lowpass, bandpass, and / or highpass filters, along with down sampling of the filtered signals [7].If certain conditions are meeting the design of the filters, then perfect reconstruction of the original signal can be obtained by using a reconstruction scheme of upsampling 1034 and filtering to remove spectral aliasing effects.In this study, the Discrete Wavelet Transform (DWT) is used to implement subband decomposition and reconstruction filter banks .The DWT can be flexible and extended to work in two and higher dimensions by using separable filters working on separate dimensions.An excellent description of a twodimensional wavelet filtering scheme can be found in [8,9].In this paper, the operations of computing the forward and inverse DWT, regardless of dimension, will be denoted DWT and IDWT, respectively.It is possible to state that the most important features of a watermarking technique are that the watermark unnoticeable, robust and blind, i.e., the watermark decoder must not require the original image for extracting the embedded code.The algorithm below explains how an image is loaded and how it is divided by DWT, see the flowchart in Fig. (11).The technique has the ability to locate the region that has been altered.In this paper, watermarks have been embedded into images using wavelet packets in order to fulfill the above characteristics.The new technique (meaning by factoring) is distributed in three bands (LH,HL,HH).The following are the details of these techniques see Fig. (12).

Back Propagation Neural Network
The backpropagation (BP) algorithm is also known as error backpropagation or back error propagation or the generalized delta rule [10].The networks that get trained like this are sometimes known as multilayer perceptrons or MLPs.

Start
In the construction used for back propagation network there are nine input layers, fifteen nodes for the first hidden layer and five nodes for the second hidden layer and one output layer.The input's are called the (neighborhood pixels) wich lies around watermarking Rijndael code, the hidden contain many nodes to train the network and the output to give the actual value.The weights interconnecting the input and the hidden nodes are adjusted and learned, the weights interconnecting the hidden and the output nodes are also trained and adjusted to reach the actual value [11].
Backpropagation algorithm has been used to recognize and detect watermarking Rijndael cipher code.Binary digit code for the watermarking string Rijndael code has been decoded.One of the critical issues in watermarking Rijndael code is the feature selection for the back propagation layers construction and is dependent on the choice of the features for each case used [11,12].The aim was to recognize all bits for watermark Rijndael cipher code and the goal to create a network that could recognize the inverse cipher to obtain the watermarking characters correctly though there were some attacks like noise and others which make errors in any bit but increase the number of copies for watermarking Rijndael which will defeat any attacking appearing in image.

Figure 5 A
Figure 5 A block diagram for receiving image after applying typical extraction method using BPNN.

Figure 7
Figure 7 Example of a state with Nb=6 and cipher key with Nk=4

Fig. 8
Fig. 8 illustrates the effect of the ShiftRow transformation on the State.

Figure 8
Figure 8 The effect of ShiftRow Transformation

Table( 2 )
Shift offsets for different block length 1033

Figure 11
Figure 11 Flowchart illustrate DWT used to distribute watermarking Rijndael Code.

Figure 12
Figure 12 Applying discrete wave let transform and distributed watermarking technique.

Figure 17 Figure 16
Figure 17 The relation to calculate MSE for BPNN training network

1
The received invisible image with watermarking Rijndael cipher code technique is shown in Fig.21, while Table(4) shows the calculations of epoch and MSE.

Figure
Figure 19 Two dimensional original image

Figure 22
Figure 22 The relation to calculate MSE for training network