New Strongly Robust DWT Based Watermarking Algorithm

Essay add: 22-10-2015, 20:34   /   Views: 96

Abstract- In this paper we have presented two watermarking algorithms. First one is a new strongly robust scheme for copyright protection. This scheme is based on 'Discrete Wavelet Transform', by embedding scrambled watermark in HL subband at level 3. Direct weighting factor is used in watermark embedding and extraction process.

This scheme results in exact recovery of watermark with standard database images of size 512x512, giving Correlation Factor equals to 1. The Correlation Factor for different attacks like Noise addition, Filtering, Rotation and Compression ranges from 0.90 to 0.95. The PSNR with weighting factor 0.02 is up to 48.53 dBs. This is nonblind and embeds binary watermark of 64x64 size.

The second technique is traditional method of watermarking. We also tried to compare advanced scheme of first type with traditional method and recommended our advanced scheme.Keywords-DWT, Scrambling, Arnold Transform,Copyright.


It has become a daily need to create copy, transmit and distribute digital data as a part of widespread use of multimedia technology in internet era. Hence copyright protection has become essential to avoid unauthorized replication problem. Digital image watermarking provides copyright protection to image by hiding appropriate information in original image to declare rightful ownership [1].

Robustness, Perceptual transparency, capacity and Blind watermarking are four essential factors to determine quality of watermarking scheme[4][5]. Watermarking algorithms are broadly categorized as Spatial Domain Watermarking and Transformed domain watermarking. In spatial domain, watermark is embedded by directly modifying pixel values of cover image. Least Significant Bit insertion is example of spatial domain watermarking.

In Transform domain, watermark is inserted into transformed coefficients of image giving more information hiding capacity and more robustness against watermarking attacks because information can be spread out to entire image[1]. Watermarking using Discrete Wavelet Transform, Discrete Cosine Transform, CDMA based Spread Spectrum Watermarking are examples of Transform Domain Watermarking. The rest of the paper is organized as follows: Section II focuses on survey of existing digital image watermarking algorithms. Section III focuses on importance of Discrete Wavelet Transform.

In section IV, we have presented two watermarking schemes: In first scheme a new strongly robust DWT based algorithm is presented and second scheme is traditional technique. Section V shows Experimental results after implementation and Testing for both schemes. In section VI, we have concluded and recommend our firstly DWT based scheme.


In traditional watermarking approach some LSB based as well as watermarking methods with pseudo random generator are proposed [3]. In transform domain methods, watermarking using CWT, only DWT, only DCT or combined approach of DWT-DCT are proposed. In CWT, Calculating wavelet coefficients at every possible scale is huge amount of work, and it generates a lot of data. There is highly redundant information as per as the reconstruction of the signal is concerned. Due to the attractive features of Discrete Wavelet Transform, researches have been focused on DWT [15].

Wang Hongjun, Li Na have proposed a DWT based method [14] in which watermark was embedded in middle frequency coefficient using α as flexing factor with α =β |m| ,where m is mean value of all coefficients watermarking embedded. But this method doesn't provide enough security. The method proposed in [14] using DWT was extended in [15] to enhance security of algorithm by using Arnold's Transform pretreatment for watermark.

But this method can be extended to improve PSNR and security levels. As given in [16], two phase watermark embedding process was carried out using DWT. Phase 1: Visible watermark logo embedding, Phase 2: Feature extracted watermark logo embedding. The algorithm was based on Texture Based Watermarking.

A Integer Wavelet Transform with Bit Plane complexity Segmentation is used with more data hiding capacity. [2]. But this method needs separate processing for R, G and B components of color image. As given in [17] using DWT, host image is decomposed into 3 levels recursively.

In level one we get 4 sub bands. In level 2, each subband of level 1 is divided to 4 sub bands to give total 16 sub bands. Finally, each subband of level 2 is again divided into 4 sub bands each to give total 64 sub bands. Then' Generic algorithm' was applied to find the best subband for watermark embedding to provide perceptual transparency and robustness.

But the process is too lengthy and time consuming. The common problem with DCT watermarking is block based scaling of watermark image changes scaling factors block by block and results in visual discontinuity.[1][6]. As given in [13], J. Cox et. al had presented 'Spread spectrum based watermarking schemes', Chris Shoemaker has developed.


DWT has become researchers focus for watermarking as DWT is very similar to theoretical model of Human Visual System (HVS). ISO has developed and generalized still image compression standard JPEG2000 which substitutes DWT for DCT. DWT offers mutiresolution representation of a image and DWT gives perfect reconstruction of decomposed image.

Discrete wavelet can be represented as(1)For dyadic wavelets a0 =2 and b0 =1, Hence we have,j, k (2)Image itself is considered as two dimensional signal. When image is passed through series of low pass and high pass filters, DWT decomposes the image into sub bands of different resolutions[11][12]. Decompositions can be done at different DWT levels.Fig 1: Three Level Image DecompositionAt level 1, DWT decomposes image into four nonoverlapping multiresolution sub bands: LLx (Approximate sub band), HLx (Horizontal subband), LHx (Vertical subband) and HHx (Diagonal Subband).

Here, LLx is low frequency component whereas HLx, LHx and HHx are high frequency (detail) components [7][8][9].To obtain next coarser scale of wavelet coefficients after level 1, the subband LL1 is further processed until final N scale reached. When N is reached, we have 3N+1 subbands with LLx (Approximate Components.) and HLx, LHx, HHx (Detail components) where x ranges from 1 to N. Three level image decomposition is shown in Fig :1. Embedding watermark in low frequency coefficients can increase robustness significantly but maximum energy of most of the natural images is concentrated in approximate (LLx) subband. Hence modification in this low frequency subband will cause severe and unacceptable image degradation. Hence watermark is not be embedded in LLx subband.

The good areas for watermark embedding are high frequency subbands (HLx, LHx and HHx), because human naked eyes are not sensitive to these subbands. They yield effective watermarking without being perceived by human eyes. But HHx subband includes edges and textures of the image. Hence HHx is also excluded.

Most of the watermarking algorithms have been failed to achieve perceptual transparency and robustness simultaneously because these two requirements are conflicting to each other. The rest options are HLx and LHx. But Human Visual System (HVS) is more sensitive in horizontal than vertical. Hence Watermarking done in HLx



This scheme is improvement of algorithm presented in 2008 by Na Li et. al, given in [15] using Discrete Wavelet Transform with Arnold Transform. The improvement is made in following aspects: The security level is increased by introducing "PN Sequence' depending on Arnold periodicity and depending on threshold value absolute difference of Arnold Transformed-Watermark-images is embedded. Instead of calculating flexing factor related to mean value of coefficients of watermark image, here directly appropriate weighting factor is selected.

The Image decomposition is done with 'Haar' which is simple, symmetric and orthogonal wavelet.Watermark Scrambling:Watermark Scrambling is carried out through many steps to improve security levels. Different methods can be used for image scrambling such as Fass Curve, Gray Code, Arnold Transform, Magic square etc. Here Arnold Transform is used. The special property of Arnold Transform is that image comes to it's original state after certain number of iterations.

These 'number of iterations' are called 'Arnold Period' or 'Periodicity of Arnold Transform'. The Arnold Transform of image is(3)Where, (x,y) ={0,1,.....N} are pixel coordinates from original image.(, ) :corresponding results after Arnold Transform.Periodicity of Arnold Transform:The periodicity of Arnold Transform (P), is dependent on size of given image. From equation: 3 we have,(4)(5)If (mod (, N) ==1 && mod (, N) ==1)then P=N (6)Embedding Algorithm:Step 1: Decompose the cover image using simple 'Haar' Wavelet into four nonoverlapping multiresolution coefficient sets: LL1, HL1, LH1 and HH1.Step 2: Perform second level DWT on LL1 to give 4 coefficients: LL2, HL2, LH2 and HH2.Step 3: Repeat decomposition for LL2 to give next level components: LL3, HL3, LH3 and HH3 as shown in fig 1.Step 4: Find Arnold periodicity 'P' of watermark using equation 6.Step 5: Determine 'KEY' where. Then generate PN Sequence depending on 'KEY' and find the sum of random sequence say SUM.Step 6: If SUM > T where, T is some predefined Threshold value, then find two scrambled images applying Arnold Transform with KEY1 and KEY2, where, ,

, .

Now, Take absolute difference of two scrambled images to give 'Final Scrambled image'.Step 7: If SUM <T, then apply Arnold Transform directly to watermark image with 'KEY' to get 'Final Scrambled image'.Step 8: Add 'Final Scrambled image' to HL3 coefficients of cover image as follows:(7)Where, K1 is weighting factor, New_HL3 (i, j) is newly calculated coefficients of level3, Watermark (i, j) is 'Final Scrambled image'.Step 9: Take IDWT at Level3, Level2 and Level1 sequentially to get 'Watermarked Image.Extraction Algorithm:The proposed method is nonblind. Hence the original image is required for extraction process. The simple algorithmic steps are applied are given below.Step 1: Decompose Cover image using 'Haar' wavelet up to 3 levels to get HL3 Coefficients.Step 2: Decompose 'Watermarked Image' using 'Haar' wavelet up to 3 levels to get HL3'.Step 3: Apply Extraction formula as follows:(8)IfOtherwiseStep 4: Perform 'Image Scrambling' using 'Arnold Transform' with' KEY' that we had used in embedding process to recover the Watermark.Fig: 2 Watermark EmbeddingFig: 3 Watermark Extraction


This spatial domain, watermarking is traditional scheme of watermarking. Here watermark is embedded by directly modifying pixel values of cover image as given below.Watermark EmbeddingStep 1. Read gray scale Cover Image and Watermark.Step2.Consider binary of pixel values of Cover Image and make it's n Least Significant Bits 0e.g. For n=4, Binary of 143=>10001111 and Making 4 LSB 0 =>10000000=>128 is decimal equivalent.Step: 3 Consider binary of pixel values of Watermark and right shift by k bits where k=8-n. For n=4, k will be 4. Binary of 36=>100100 and after right shift by 4: 000010=>2 is decimal equivalentStep 4: Add result of step 1 and step 2 to give watermarked image. E.g.

Add 128+2=>130. This gives pixel value of watermarked image=>10000010Fig: 4 Pixel of Cover image (Original Image), Watermark,Watermarked Image and Extracted WatermarkWatermark Extraction:Take pixels of watermarked Image and left shift by k bits where k=8-n. e.g. Left shift by 4=>00100000 =>32 . This gives pixels of Extracted Watermark.

The sample values of Pixel of Cover image, Watermark,Watermarked_Image and Extracted Watermark are shown in fig.4.


Results of Scheme- 1:The project is implemented in Matlab and standard database images with 512x512 sizes as cover image and 64x64 size binary watermark images are used for testing. The performance Evaluation is done by two performance evaluation metrics: Perceptual transparency and Robustness.Perceptual transparency means perceived quality of image should not be destroyed by presence of watermark. The quality of watermarked image is measured by PSNR. Bigger is PSNR, better is quality of watermarked image. PSNR for image with size M x N is given by:(9)Where, f (i, j) is pixel gray values of original image. f '(i, j) is pixel gray values of watermarked image.MaxI is the maximum pixel value of image which is equal to 255 for gray scale image where pixels are represented with 8 bits.

Robustness is measure of immunity of watermark against attempts to remove or destroy it by image modification and manipulation like compression, filtering, rotation, scaling, collision attacks, resizing, cropping etc. It is measured in terms of correlation factor. The correlation factor measures the similarity and difference between original 'watermark and extracted watermark. It' value is generally 0 to 1. Ideally it should be 1 but the value 0.75 is acceptable.

Robustness is given by:(10)Where, N is number of pixels in watermark, wi is original watermark, wi' is extracted watermark.Fig 5 (a) Cover Image (b) Watermarked Image(c) Recovered WatermarkHere, we are getting PSNR 48.53 dB and =1, for weighting factor K1=0.02. The PSNR and for 'standard database images' with coeresponding test image and recovered watermarks are shown in Table 1. The gray scale 'lena' image is tested for various attacks given in Table 2. Here, we are getting within range of 0.90-0.95 for various attacks. This shows that 'watermark recovery' is satisfactory under different attacks.Table 1: Experimental results for standard database images with size 512x512Table 2: Experimental results for various attacks withK1=0.07, 'Lena' image, size 512x512Results of Scheme- 2:This algorithm has simple implementation logic.

We have tested with PSNR less than 23 for different attacks as shown in figure 6.Fig: 6: Experimental results with PSNR for NoiseAttacks with various intensities.


First scheme presented here is a new strongly robust 'Digital Image Watermarking' with increased security levels and producing exact recovery of original watermark for standard image database, giving correlation factor equals to 1 and PSNR up to 48.53 dBs. Experimental results have demonstrated that, this technique is very effective supporting more security. As per ISO's norms, the still Image Compression standard JPEG2000 has replaced Discrete Cosine Transform by Discrete Wavelet Transform.

This is the reason why more researchers are focusing on DWT, which we have used for implementation. The presented 'Digital Image Watermarking' methodology can be extended for 'color images and videos' for authentication and copyright protection. Hence we are strongly recommending our DWT based scheme which is presented here.


We are thankful to BCUD, University of Pune for providing 'Research Grant' for the project "Transformed based strongly Robust Digital Image Watermarking" in academic year 2010-2011.

Article name: New Strongly Robust DWT Based Watermarking Algorithm essay, research paper, dissertation