(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, o. 7, 2010
m jik k ji
(1)Equation (1) indicates that each bit of the watermark atlocation (i, j) will be moved to a new location (i*, j*).Parameters k and m are secret keys needed for disarranging and rearranging the watermark bits. These permuted watermark bits will be embedded bit by bitand line by line into the host image.
: Arrange the host image blocks sequentially one by one and row by row. Use a pseudo-random number generator to determine the sequence of host image blocks used for embedding the permuted watermark bits. A pseudo-random number generator with a givenseed always generates the same sequence of randomnumbers. Thus, the seed of the pseudo-random number generator is a secret key needed to embed and extractthe permuted watermark bits. Steps 2 and 3 offer cryptographic protection against intentional attacks because the secret keys of the pseudo-random number generator and TA permutation are necessary for extracting and rearranging the embedded watermark for subsequent destruction.
: DCT-transform the next host image block fromthe spatial domain to the frequency domain. Each block is DCT-transformed using the following equation :
j y i x x y y x pixel jC iC ji DCT
ji for otherwise jC iC
In equation (2), DCT (i, j) represents the value atlocation (i, j) in the DCT-transformed block, while pixel(x, y) represents the value at location (x, y) in theoriginal block. N is the number of locations (pixels) ineach block.
: Embed the next permuted watermark bit into thelow frequencies of the DCT-transformed block. Toincrease the robustness of the embedded watermark against unintentional attacks, as explained in Section II,the locations for embedding the permuted watermark bitare restricted to the low frequency locations 0, 1, 2, and3 as shown in Figure 1. The embedding location withthe largest DCT coefficient is selected as the embeddinglocation for a permuted watermark bit with value 1,while the embedding location with the smallest DCTcoefficient is selected as the embedding location for a permuted watermark bit with value 0. It should be notedthat the largest and the smallest values of the DCTcoefficients are relative values that are unlikely tochange under attacks. This technique, thus, attempts toutilize the more robust relative values of the DCTcoefficients instead of the absolute values that can easilychange under attacks. It should be also noted that we donot need to physically embed the permuted watermark bit. We only need to remember its embedding location.This implies that in the proposed technique theembedded watermark is 100% transparent. Since thereare only 4 possible embedding locations, we need onlytwo bits to represent each embedding location. For example, (0, 0), (0, 1), (1, 0), and (1, 1) can be used torepresent embedding locations 0, 1, 2, and 3respectively. Thus, the permuted watermark bit isreplaced by two bits representing its embeddinglocation. Steps 4 and 5 are repeated until all the permuted watermark bits have been embedded in theDCT-transformed host image blocks. Since each permuted watermark bit is represented by its embeddinglocation and since an embedding location is represented by two bits, the final result is an array with double thesize of the watermark array.
: After embedding all the permuted watermark bits into the DCT-transformed host image blocks, theresultant array of the embedding locations of the permuted watermark bits is decomposed into two arrayswith the same size. The first array stores the first bitrepresenting each permuted watermark bit embeddinglocation, while the second array stores the second bit.The two arrays are of the same size as the watermark array and are considered to be the watermark shares or transparencies . For example, suppose that theembedding location of the permuted watermark bit (a, b)is location 1 as shown in Figure 1. This embeddinglocation is represented by (0, 1). Thus, thecorresponding value of the bit (a, b) in share or transparency 1 is 0 and in share or transparency 2 is 1.One of the generated transparencies or shares is public,while the other is secret and is registered to the CertifiedAuthority (CA) for additional security and protectionagainst intentional attacks (that attempt to extract theembedded watermark for subsequent destruction) asmentioned before.
: Finally, inverse DCT-transform eachwatermarked block from the frequency domain to thespatial domain forming the watermarked host imageusing the following equation :
j y i x ji DCT i j jC iC y x pixel