FORM - 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENTS RULES, 2003
PROVISIONAL
Specification
(See section 10 and rule 13)
METHOD AND APPARATUS FOR WATERMARKING
TATA CONSULTANCY SERVICES LIMITED
an Indian Company
of Bombay House, 24, Homi Mody Street, Mumbai 400 001,
Maharashtra, India
THE FOLLOWING SPEC IFICATION DESCRIBES THE INVENTION.

Field of the Invention:
This invention relates to video coding decoding (compression decompression).
In particular, this invention envisages a novel way of embedding watermark during video sequence based on block based coding like MPEG series and H.261.
Still particularly, this invention relates to an H.264 video CODECs (Encoder-Decoders) and in particular envisages a method for watermarking technique applied in transformation and quantization detection module.
Background of the Invention:
Last few years have witnessed rapid growth in video coding technology. Among various standards, H.264/Advanced Video Codec (AVC) is found to be of significant importance regarding reduced bandwidth, better image quality and network friendliness. With the advancements in high speed and high bandwidth network, lots of web based multimedia applications have evolved that enables transmission of multimedia data in real time over Internet. As an effect of these advancements, the video storage and distribution industry faces a major problem in copyright protection and authentication, which hampers the business of multimedia contents, too. One of the current fields of interest is to develop a system with authentication and copyright protection methodology embedded within an efficient video codec. The objective of this work is to devise a watermarking algorithm that works in
2

the H.264 compressed domain that is robust and at the same time is of very low computational complexity to meet the real time criteria.
High-speed computer networks, the Internet and the World Wide Web have revolutionized the way in which digital data is distributed. The widespread and easy accesses to multimedia contents and possibility to make unlimited copy without loss of considerable fidelity have motivated the need for digital rights management. Digital watermarking is a technology that can serve this purpose.
A watermark is a digital data embedded in multimedia objects such that the watermark can be detected or extracted at later times in order to make an assertion about the object. The main purpose of digital watermarking is to embed information imperceptibly and robustly in the host data. Typically the watermark contains information about the origin, ownership, destination, copy control, transaction etc. Potential applications of digital watermarking include transaction tracking, copy control, authentication, legacy system enhancement and database linking etc.
Growing popularity of video based applications such as Internet multimedia, wireless video, personal video recorders, video-on-demand, set-top box, videophone and videoconferencing have a demand for much higher compression to meet bandwidth criteria and best video quality as possible. Different video Encoder Decoders (CODECs) have evolved to meet the current requirements of video application based products. Among various available standards H.264 / Advanced Video Codec (AVC) is becoming an important alternative regarding reduced band width, better image quality in
3

terms of peak-signal-to-noise-ratio (PSNR) and network friendliness, but it requires higher computational complexity.
A large number of watermarking schemes have been proposed to hide copyright marks and other information in digital images, video, audio and other multimedia objects. Different watermarking techniques have been proposed for different video CODECs, but only a few works on H.264/AVC can be found in the literature. H.264/AVC uses different transformation and block sizes than MPEG series, so development of new algorithms is required to integrate robust watermarking techniques for different profiles of H.264/AVC. Since 1990 several papers were published on "robust watermarking technique". But most of these techniques are not robust. Most of the existing methods of watermarking splits the video sequence into frames and apply the image watermarking technique in each frame. The problem of these types of approaches is that they never consider the additional temporal information available in video sequence. Moreover these approaches cannot preserve the security against frame dropping or frame averaging attacks. Over and above these techniques did not exploit the architecture of video codec that can leads to a robust encryption technique with a very less amount of computational cost.
Of the works on watermarking technique in H.264, most of them embed information in motion vectors. This technique creates problem for the video sequence with high motion and zero motion sequences. Moreover this technique did not propose any method for checking integrity, i.e. the method for checking whether the original video sequence gets tampered or not.
4

In the system proposed in accordance with this invention, a watermark is embedded by exploiting the video encoding architecture that avoid the problem of image watermarking technique applied in video sequence. Thus the method in accordance with this invention is robust against frame averaging and frame dropping attacks. The IP address, time stamp and company logo is typically used as the embedding message. Moreover the entire stream is hashed to produce a unique number and that is again embedded as the watermark in some other frame to check the integrity.
Therefore this invention envisages embedding information, typically using hashing in watermarking.
Further the invention envisages a method of modifying DCT coefficients.
Further the invention envisages applying the method of this invention to H.264 AVC based video encoders, without sacrificing significantly on the video quality and having significantly less computational complexity.
The method can be extended in application to other video CODEC that may use similar block based approach and header format.
Brief Description of the Accompanying Drawings:
The invention will now be described with reference to the accompanying drawings in which;
Figure 1 shows the Level 1 Block diagram of watermarking process;
5

Figure 2 shows the Block diagram of logo size checking process;
Figure 3 shows the Block diagram of Position selection;
Figure 4 shows the Degradation of image quality due to watermarking;
Figure 5 shows Table 1: Results showing computational complexity of
Watermark algorithm;
Figure 6 shows Table 2: Image quality with and without watermarking at
different bitrates;
Figure 7 shows Table 3: List of Acronyms used; and
Figure 8 shows the Table 4: Conclusion
Detailed Description of Invention
In the watermarking method proposed in accordance with this invention, watermarking in H.264 encoder is achieved in two folds: in even numbered Independent Decoder Refresh (IDR) frames, three different messages are embedded and in odd numbered frames the bitstream is obtained by hashing the last Group of Picture (GOP) to ensure integrity. The process is described in details as follows:
The algorithm of watermark embedding is described below. It is also depicted in pictorially.
• Is the frame an even numbered IDR?
o If Yes, embed logo, timestamp and IP address or key.
o Else Hash the last GOP and embed the number.
Test for watermarking message size:
6

In even numbered IDRs three messages are added: logo of the company in
binary format, timestamp of the encoding time and the IP address or any 32 bit
key value of the source end. These messages are embed into the even
numbered IDRs using the following method:
For logo, the information is embedded in diagonal sub-blocks only. First the
logo size should be less than the number of diagonal sub-blocks. If the video
sequence is considered with height h and width w., it can be assumed that the
macroblocks (MB) are of size 16 x 16 and sub-macroblocks (SB) are of size 4
x4.
So, Total number of macroblock = (h/16) x (w/16) = TotMBnr (say)
Number of sub-block in a macro-block = (16 x 16)/(4x4)= 16
Total Number of sub-block = TotMBnr * 16
Total number of diagonal sub-block in a MB = 16/4 = 4
Total number of diagonal sub-block in the sequence = (TotMBnr * 16) / 4 =
TotMBnr * 4
This allowed number of bits can be defined as BitsAllowed.
Resolution of binary logo file'<= BitsAllowed.
So in a nutshell, the number of binary values (or bits) in the logo files should
be less than or equal to four times of the total number of MB.
Timestamp is taken a four-byte number and the IP address or source key is
another 4 byte number.
This process of checking the size is depicted in Figure 2.
Find the location for embedding Watermark:
Selection of the position of DCT coefficient to be modified is based on our
following observations:
7

• DCT coefficients are zero in most cases
• Most significant information lies in top and left
• Modification of diagonal elements at right and bottom results in insignificant artifacts
• Coefficients in diagonal positions are more stable than the others.
Embed one watermarking bit in one diagonal coefficient of diagonal sub-block
For Binary image like Logo:
• Find whether the sub-block is diagonal
• Every sub-block has 16 coefficients (4x4)
• Embed the watermark in 10th or 15th coefficients only
o If the bit number to be embedded is odd add it in 10th coefficient
o Else in 15th Coefficient
For text message like Timestamp and IP address:
• Find whether the sub-block is ab-diagonal
• Embed the watermark in 10th or 15lh coefficients only
o If the bit number to be embedded is odd add it in 10th coefficient
o Else in 15th Coefficient
This process is described in Figure 3.
The algorithm of watermarking:
Two basic intra-prediction modes INTRA-4x4 and INTRA-16xl6 are used in H.264/AVC, comprising 4x4 and 16 x 16 block-wise directional spatial predictions, in which the pixel value of the current block is predicted by the
8

edge pixels of the adjacent blocks. Then, the prediction error is transformed primarily by a new 4x4 integer DCT instead of the float 8x8 DCT, which is widely used in existing standards. While the smaller block-size is justified by the advance of prediction capabilities by using above mentioned prediction modes, it makes the embedded watermark more sensitive to attacks or transcoding.
Watermark bits are inserted by altering the quantized AC coefficients of
luminance blocks within I-frames. In order to survive the re-compression, two
major obstacles are considered as follows.
First of all, the watermark signal M(u, v) must be strong enough to survive the
quantization, so that
|Mq(u,v)| = |quant[M(u,v),QP]| (1)
Where the quant [•] denotes the quantization operation, QP denotes the quantization parameter and (u, v) denotes a position in a 4 M block BK. Obviously M(u,v) should be even greater if the watermark is required to survive the re-quantization during transcoding.
Furthermore, since the change of the prediction direction (or mode) during transcoding may alter the value of DCT coefficients and thus leads to watermark detection error, one of the quantized AC coefficients Xq(u, v) is
chosen in high frequency along the diagonal positions (i.e., u = v) for embedding. The Xq(u, v) is replaced by the watermarked coefficient Xq,

9

Where wn is the bit to be embedded. It is noticed that the AC coefficient Xq(u, v) is cleared if '0' is embedded. It can be justified by the fact that the Xq(u, v) is zero in most cases. It will not introduce significant artifacts. After watermarking, the best mode for a watermarked macroblock S.k is
selected by minimizing the modified Lagrange optimization function:
Where DRECand RREC represent the distortion and the number of bits, respectively, encoded for modes
is Lagrange parameter.
Results and discussion:
The time complexity of the watermarking method in accordance with this invention is seen in terms of basic arithmetic operation like ADD, MULTIPLY, DIVISION, MODULO in Figure 5. Total cycles can be simulated from that figures. Figure 4 depicts the variation of PSNR values per frame for Y, U and V components of test stream akiyo.yuv. As it can be readily observed from the figure, the maximum PSNR variation in any component is less than 3 dB, which signifies that the original video and watermarked video is visually indistinguishable. Figure 6 shows the variations and percentage degradation of PSNR after watermarking in foman.yuv test sequences. The percentage degradation shows the effectiveness of the watermarking algorithm.
Figure 7 describes the list of acronyms used in the tables, whereas table 4 tabulates the video quality, retrieved image quality and retrieved text quality after attack.
10

Conclusion: From the above results it can be concluded that the water marking technique in accordance with this invention is a robust watermark technique with very less complex algorithms. Figure 8 clearly shows that the watermarking technique in accordance with this invention cannot preserve the image quality only if the video quality drastically degraded during attack. If the attack is good enough to produce acceptable video quality after attack, the embedded logo and the text are retrieved without any error. If the attack degrades the image to a poor quality it is of no practical implication, as no one will opt for video of this quality. In the above results, it can be seen that in averaging attack and resize attack the video quality is maintained. Again the retrieved image and text messages are retrieved without any error for these cases. That proves the robustness of the watermarking scheme in accordance with this invention.
Although the invention has been described in terms of particular embodiments and applications, one of ordinary skill in the art, in light of this teaching, can generate additional embodiments and modifications without departing from the spirit of or exceeding the scope of the claimed invention. Accordingly, it is to be understood that the drawings and descriptions herein are proffered by way of example to facilitate comprehension of the invention and should not be construed to limit the scope thereof.

11