# Transform Domain Techniques Health And Social Care Essay

Many spatial techniques are based on adding fixed amplitude pseudo noise sequences to an image. Pseudo random noise (PN) sequences are used as the spreading key when considering the host media as the noise in spread spectrum system, where the watermark is the transmitted multimedia content. Many techniques have been proposed in the spatial domain such as LSB (Least Significant Bit) insertion method, the patch work method and the texture block coding method [6]. These techniques process the location and luminance of the image pixel directly. The LSB method has a major disadvantage that the least significant bits may be easily destroyed by lossy compression. The invisibility of the watermark is achieved on the assumption that the LSB data are visually insignificant. In general, the techniques that modify the LSB of the data using a fixed magnitude PN sequence are extremely sensitive to signal processing operations and weak to watermark attacks. The contributing factor to this weakness is the fact that the watermark must be invisible. As a result, the magnitude of the embedded noise is limited by the smooth regions of the image, which most easily exhibit the embedded noise.

## 2.3.2 Transform Domain Techniques

Transform domain method based on special transformations, and process the coefficients in frequency domain to hide the data. Transform domain methods include Fast Fourier Transform(FFT), Discrete Cosine Transform(DCT), Discrete Wavelet Transform(DWT), Curvelet Transform(CT), Counterlet Transform(CLT) etc. In these methods the watermark is hidden in the high and middle frequency coefficients of the cover image. The low frequency coefficients are suppressed by filtering as noise, hence watermark is not inserted in low frequency coefficients [7]. The transform domain method is more robust than the spatial domain method against compression, filtering, rotation, cropping and noise attack etc.

Many transform based digital image watermarking techniques have been proposed by researchers and scientists. To embed a watermark, first transform is applied on the cover image and then modifications are made to the transformed coefficients.

Cox et al [32] find parallels between spread-spectrum communications and watermarking and used a frequency domain transform to convert an image into another domain.

In frequency domain, a sequence of values I0= I0[1], I0[2], ……I0[n] are extracted from the given carrier image and then this sequence is modified as per the requirement. The watermark is a sequence real numbers w = w[1], w[2] ...…w[n]. Each value of this watermark sequence is chosen independently from the Gaussian distribution with zero mean and with variance unity.

Three different formulas to embed watermark, whose difference lies in their embedding characteristics and in their invertibility are given below:

Iw[i] = I[i] +αw[i] ………………………… (2.1)

Iw[i] = I[i] (1+αw[i]) ……………………… (2.2)

Iw[i] = I[i] +exp (αw[i]) ……………………. (2.3)

Where α is the scaling or watermark strength parameter, which influences he robustness as well as the imperceptibility of the watermarked image.

Watermarking can be implemented in frequency domain such as proposed by Cox et al [32], where the embedding technique is based on DCT and Pseudo Noise sequence. The watermark extraction is based on the knowledge of cover image and the frequency locations of the watermark. The normalized correlation coefficient is computed and set to a certain threshold. If the normalized correlation coefficient is large enough, the watermark is detected. This Cox et al method is robust to image scaling, JPEG compression, dithering, cropping, and rescanning.

Another watermarking scheme in frequency domain is wavelet transform technique. Barni et al [33], proposed a watermarking method on decomposition of wavelet transforms. The technique based on the decomposition of input cover image into low and high frequency components with different orientations. A Discrete Wavelet Transform is applied to the cover image. The watermark is inserted into the highest level subbands as per the following rule:

IwLH [i,j]= I0LH[i,j]+αβLH[i,j]w[iN+j] ………..(2.4)

IwHH [i,j]= I0HH[i,j]+αβHH[i,j]w[MN+iN+j] ……(2.5)

IwHH [i,j]= I0HH[i,j]+αβHH[i,j]w[2MN+iN+j] ……….(2.6)

Where α is the global parameter for the watermarking strength, β is the local weighting factor and w is the pseudo random binary sequence. The masking characteristics of human visual system depend on this local weighting factor. The correlation between the watermark DWT coefficients and the watermark sequence is computed to retrieve the watermark.

The similarity between the correlation method and Barni’s method is shown by Cox et al. [32]. This algorithm is formulated as a correlation by defining the pattern with the same dimensions as that of coefficient matrix. The pattern values are determined by the influence of the corresponding frequency coefficients. The pattern is zero for coefficients not considered in the evaluation. The pattern for the pair coefficients is either 1 or -1. Thus, the sign of the correlation directly depends on the relation of the pair of coefficients.

Fractal watermarking schemes are based on fractal compression, which is developed based on iterated function systems. The fractal encoding algorithm partitions the original image into non- overlapping domain cells. The image is covered with overlapping domain cells. For each range cell, the corresponding domain cell and transform are searched to determine the best cell range. This step is computationally expensive. The range of transforms typically includes affine transforms, change of brightness and contrast. This transform describes the self-similarity between range cell and the corresponding domain cell. To embed the watermark, the range cells are restricted by the encoded information. To retrieve the watermark from the watermarked block, the corresponding domain cells reveal the embedding information.

Samesh Oueslati et al, [13] proposed an adaptive image watermarking scheme based on Multi-Layer Feed forward (MLF) neural networks. In this method the host image is first decomposed into non-overlapping 8x8 blocks, and the DCT process is performed for each block. Coefficients are then selected for watermark insertion. Human Visual System (HVS) is adopted to further ensure the watermark invisibility. Then the luminance sensitivity, frequency sensitivity, texture sensitivity and entropy sensitivity are computed and used to as the inputs of the NNS. In this paper, neural networks are used to automatically control and maximum image–adaptive strength watermark [13].

Cheng-Ri Piao et al, [16] proposed a blind watermarking algorithm based on HVS and RBF neural network for digital images. In this method, RBF is implemented while embedding and extracting watermark. The human visual system model is used to determine the watermark insertion strength. The inserted watermark is a random sequence. The secret key determines the beginning position of the image where the watermark is embedded. This process prevents possible pirates from removing the watermark easily [16].

Nizar Sakr et al, [20] proposed an adaptive wavelet-based watermarking algorithm that is based on the model of a HVS and a Fuzzy Inference System (FIS).In this method; Sugeno-type fuzzy model is exploited in order to determine a valid approximation of the quantization step of each DWT coefficient. Furthermore, the HVS properties are modeled using biorthogonal wavelets to improve watermark robustness and imperceptibility [20].

## Nagaraj.V.darwadkar Method

Nagaraj V. Dharwadkar et al [6], proposed a non-blind watermarking scheme for color images in RGB space using DWT-SVD in 2010. In this method, the watermark is embedded into cover image in RGB space. The combinations of discrete wavelet transform and singular value decomposition of blue channel is used to embed watermark. The singular values of different sub band coefficients of blue channel are modified using different scaling factors to embed the singular values of the watermark. The copy of the watermark is embedded into four sub band coefficients which are very difficult to remove or destroy [6].

## Watermark embedding procedure

Step 1: Read the color image I of size NxN.

Step 2: Read the monochrome image X of size MxM and apply DWT on X to get D= {dij} of size MxM.

Step 3: Compute R, G and B channels from color image I of size NxN.

Step 4: Transform R, G and B channels into Y, I and Q channels of the color image.

Step 5: Compute third level DWT on Y channel to get the frequency components {HH1, HL1, LH1, {{HH2, HL2, LH2, {{{HH3, HL3, LH3}}}}.

Step 6: Embed the watermark frequency coefficients, starting from HH1 for each row select the frequency coefficients in descending order with respect their absolute values.

Step 7: Modify each frequency coefficient f of cover image to ij. If the subcomponent HH1 is insufficient to embed the complete watermark, then insert in the other coefficients in the order {HL1, LH1, {{HH2, HL2, LH2, {{{HH3, HL3, LH3}}}}.

Step 8: Save the location of the modified frequency coefficients into a key array K of size NxN. The key array consists of value 1 if the coefficient is modified otherwise 0.

Step 9: Replace by in decomposed y channel and compute inverse DWT of modified Y channel.

Step 10: combine modified Y channel with I and Q to get watermarked image.

## Watermark extraction procedure

Step 1: Read the watermarked image of size of size NxN.

Step 2: compute,’ and channels of the watermarked image.

Step 3: Transform these, and channels into, and channels.

Step 4: Compute third level DWT on channel to get the frequency components {HH1, HL1, LH1, {{HH2, HL2, LH2, {{{HH3, HL3, LH3}}}}.

Step 5: Compute third level DWT on Y channel of the un-watermarked image to get the frequency components {HH1, HL1, LH1, {{HH2, HL2, LH2, {{{HH3, HL3, LH3}}}}.

Step 6: Extract the watermark bits from the frequency subcomponents using the key array K as ij= (-)/α. If ij˃ T, then ij =1 other wise ij = 0.Whwere i= 1,2,3…M and j= 1,2,3…M.

## 2.3.2.2 Yanhong Zhang Method

Yanhong Zhang [7] proposed a blind watermark embedding/extracting algorithm using RBF neural network. In this method the DWT is used to overcome the blocking phenomenon problems in DCT. First, the original image is 4-scale level DWT transformed, and decided the watermarking strength according to HVS. When embedding watermark, a secret key is used to determine the watermark beginning location, and after that, embed and extract the watermark by using the trained RBF [7].

## Watermark embedding procedure

Step 1: Transform the original image using DWT as is the LH4, HL4, HH4 sub-band coefficient.

Step 2: Select the beginning position of watermark embedding coefficient using the secret key.

Step 3: Quantize the coefficient of DWT, (i+key) by Q, as the input to the RBFN and then get the output.

Step 4: Embed the watermark according to the following equation ; Where is the watermark sequence, q is quantization value and is the coefficient of the watermarked image.

Step 5: Perform IDWT to get the watermarked image.

## Watermark extraction procedure

Step 1: Transform the watermarked image by the DWT transform as with the sub band coefficients LH4, HL4, HH4.

Step 2: Quantize the DWT coefficient by Q as the input to the RBFN and then get the output.

Step 3: Extract the watermark using the following formula .

Step 4: Measure the similarity of the extracted watermark and the original watermark using the equation

Step 5: Use, threshold as a key to judge if there is an embedded watermark or not. If is larger than threshold and the location is equal to key, the watermark is affirmed.

## 2.3.2.3. He Xu, Chang Shujuan Method

He Xu, Chang Shujuan [10], proposed an adaptive image watermarking algorithm based on neural network. In this method, the watermarking signal is embedded in high frequency, which is in the lower frequency of original image by DWT joined with DCT. The ability of attracting is improved by pretreatment and retreatment of image scrambling and Hopfield network [10].

## Watermark embedding procedure

Step 1: The watermarking signal is applied as the training signal input to the Hopfield network in order to finish the storage of the watermark.

Step 2: After doing scrambling transform, the watermark signal R is generated. The affine transform is used as scrambling transform, the key is scrambling times, and then the watermark pretreatment is completed.

Step 3: The low frequency sub- image LL is extracted from the original image by using the first order DWT transform. I will be gotten by DCT transform which process 8x8 block partitioning.

Step 4: The scrambling watermark sequence is embedded in high-frequency coefficients of the image I according to the equationin order to get. Where is embedding strength in the range 0 1.

Step 5: The IDCT is performed to get the low frequency sub-image LL which contains watermark and IDWT is performed to get the watermark image.

## Watermark extraction procedure

Step 1: The detected image and original image are processed by first order DWT and T and I are gotten through DCT blocking phenomenon.

Step 2: Watermark is extracted through T and I input watermark detection module.

Step 3: The extracted watermark signal R is processed according to key inverse scrambling to get the watermark.

Step 4: The extracted watermark is applied as input to the Hopfield network and after data processing the watermark is extracted.

## 2.3.2.4. Charu Agarwal et al Method

Charu Agarwal et al, [24] proposed digital image watermarking in DCT domain using fuzzy inference system. In this method, Human Visual System (HVS) characteristics are modeled using a Fuzzy Inference System (FIS) for robust image watermarking. The fuzzy input variables corresponding to luminance sensitivity, edge sensitivity computed using threshold and contrast sensitivity computed using variance are fed to a FIS driven by ten fuzzy inference rules. The FIS produces a single output weighting factor which is used to embed a randomly generated normalized watermark with in the host image in the DCT domain. The signed image has good perceptual quality and is subject to stir mark image processing attacks. The high computed value of PSNR indicates robustness of the embedding algorithm. The watermark is extracted from the signed image using famous Cox’s algorithm [24].

## Watermark embedding procedure

Step 1: Cover image is divided into 8x8 blocks in spatial domain DCT is computed on all blocks.

Step 2: Compute edge sensitivity (threshold), luminance sensitivity and contrast sensitivity (variance) of all blocks of cover image.

Step 3: Supply these threshold, variance parameters as input to fuzzy inference system.

Step 4: Apply fuzzy inference rules to the fuzzy inference system and obtain the watermark weighting factor.

Step 5: Perform watermark embedding in low frequency DCT coefficients of cover image.

Step 6: Compute the IDCT to obtain the watermarked image.

## Watermark extraction procedure

Step 1: Compute DCT of all 8x8 blocks of cover and watermarked (signed) images.

Step 2: Subtract the computed coefficients of original image from watermarked image.

Step 3: Recover the watermark using fuzzy inference system.

Step 4: Compare the recovered watermark with the original watermark using Sim(X, X*) parameter.

## 2.3.2.5. Sameh Oueslati et al Fuzzy Method

Samesh Oueslati et al [25], proposed a fuzzy watermarking system using the wavelet technique for medical images. In this method, an adaptive watermarking algorithm performed in the wavelet domain is proposed which exploits a human visual system (HVS) and a fuzzy inference system (FIS). HVS is adopted to further ensure the watermark invisibility. The FIS is utilized to compute the optimum watermark weighting function that would enable the embedding of the maximum energy and imperceptible watermark. For the purpose of security and robustness, a watermark sequence is embedded by selectively modifying the middle- frequency parts of the image [25].

## Watermark embedding procedure

Step 1: Input the cover image and watermark image.

Step 2: Convert the watermark into a stream of binary data consisting of zeros and ones.

Step 3: Decompose the host image using Haar wavelet transform.

Step 4: Insert the data into wavelet coefficients, which have the largest values in middle frequency coefficients.

Step 5: Perform the inverse Haar wavelet transform to get the watermarked image.

Step 6: Display the watermarked image.

## Watermark extraction procedure

Step 1: Input the watermarked image.

Step 2: Decompose the watermarked image using Haar wavelet transform.

Step 3: Select the wavelet coefficients which have largest values in middle frequency sub band.

Step 4: Compare the coefficients of cover image and watermarked image depending upon the location.

If the coefficient of embedding˃ original coefficient then the data store in it is 1

If the coefficient of embedding˂ = original coefficient then the data store in it is 0

Step 5: Display the recovered image.

## 2.3.2.6. Ming-Shing Hsieh Method

Ming- Shing Hsieh [26] proposed perceptual copyright protection using multi-resolution wavelet- based watermarking and fuzzy logic. In this method, an efficiently DWT-based watermarking technique is proposed to embed signatures in images to attest the owner identification and discourage the unauthorized copying. This technique is based on utilizing a context model and fuzzy inference filter by embedding the watermarks in the larger entropy coefficients of coarser DWT sub bands [26].

## Watermark embedding procedure

Step 1: Sort the grey levels of watermark of size ‘n’ in ascending order to generate the sorted watermark.

Step 2: Decompose the host image into three levels with ten subbands of wavelet pyramid structure and choose a subband (HL3) to embed watermark.

Step 3: Calculate the weighted entropy of coefficients.

Step 4: Let the preset interval be and let t be the number of referenced coefficients used as a key to extract watermark without the host image. Coefficients with larger entropy are chosen from subband Where. The larger entropy coefficients make the watermark more robust and transparent. If then otherwise Where is used to get integer part of its argument. Let {} be the set of referenced coefficients and the coefficients to be embedded watermarks; {} is called the alternative coefficients. Sorting {} to generate {} called the sorted alternative coefficients.

Step 5: Quantize {} using a preset interval, which will extract the watermark W without the cover image.

Step 6: embed watermark SW into subband HL3 by using the equation , To+T1+T2)/3=EnixT1.

Step 7: Save the symbol of embedded subband and perform IDWT to get the watermarked image.

## Watermark extraction procedure

Step 1: Decompose watermarked image into three levels with ten subbands using DWT.

Step 2: Restore the scaling factor vi the symbol of embedded subband, symbol map of SCi, corresponsive map of Ci and SCi and corresponsive map of Wi and SWi.

Step 3: Extract the sorted watermarks by the proposed extracting watermarking algorithm.

Step 4: Rearrange the watermarks from corresponsive map of Wi and SWi to get the extracted watermark.

## 2.3.2.7. Soheila et al Method

Soheila Kiani et al [27], proposed Fractal based digital image watermarking using fuzzy C-mean clustering. In this method a new watermarking method is used to embed a binary watermark in to an image. The proposed method uses a special type of fractal coding that its parameters are contrast scaling the mean of rage block. Also, it utilizes the fuzzy C-mean clustering to address the watermark bits [27].

## Watermark embedding procedure

Step 1: The fractal encoding is applied on the original image to produce fractal codes for all range blocks.

Step 2: Apply the fuzzy C-mean clustering on all the blocks and classify them into four groups.

Step 3: As per the centers calculated in previous step determine class A and B.

Step 4: For each bit of watermark: