Professional Documents
Culture Documents
Authorized licensed use limited to: Moradabad Institute of Technology. Downloaded on December 20, 2008 at 01:31 from IEEE Xplore. Restrictions apply.
'. . _ i
V_am _I. V 3 Error Correcting Codes (ECC) for
video watermarking
In the proposed video watermarking system we have
IDC
(I )DC (I tiv crse) Di*scrtosl
tsineTlranis lrin used the Error Correction Codes (ECC) [4][5] to
IQ
M(MC QMu
Inverse illisatio
W
N1'13
VL-Ci
Mo?tinl-compensatced Predlition
= Var-iahlc-lenL(thl Ci>idci-
improve the robustness of the system. In this section
Nte'l) ~ ~<hVII ,iVah-iablc- = lcnL}1k Dccoder
we investigate the effectiveness of using different error
correcting codes in protecting codes of the watermark
Figure 1 The block diagram of watermark embedding information.
system with MPEG-2 encoder
Three Error Correcting Codes [4][5] BCH(31, 8),
Turbo (3, 1) and Conv (2,1,3) were implemented and
In the proposed watermarking scheme the tested. The testing watermark data used is plain text
watermark is embedded into each I-frame of video "abcdefgh". Using 1 byte/8 bits to represent each
sequence bit by bit when encoding the image character, the binary sequence for the watermark is
sequence to MPEG-2 video. This watermarking
01100001,01100010,01100011,01100100,
process does not alter the motion vector information 01100101,01100110,01101111,01101000
or any of the critical side information. During the The length of the binary sequence is 64 bits,
MPEG-2 video encoding process in Figure 1, each containing eight 8-bit symbols.
8 X 8 block of pixels of the I frames are transformed 3.1 Implementation of ECC Algorithms
into 8 X 8 block of DCT coefficients and then
quantized. For each quantized DCT coefficient block, We have implemented and compared three error
one bit of the watermark is embedded in the least correcting codes. The error correcting capacity of
significant bit (LSB) of the quantized DC coefficient these three ECC algorithms will be analysed in this
by changing the value of the LSB to the value of the section. The implementation of these three ECC
embedded bit. algorithm: BCH(31,8), Turbo(3,1) and Conv(2,1,3)
are described as follows:
2.2 Watermark Extracting Process
The watermark extraction process is the inverse of 3.1.1 BCH(31, 8)
watermark embedding process. When decoding
MPEG-2 video to image sequence, the watermark is Each 8-bit symbol of the watermark data is
retrieved bit by bit from the LSB of DC coefficient of individually BCH encoded. BCH(3 1, 8) is a shortened
each DCT coefficient block before de-quantization. BCH(3 1, 11) code over GF(25), which can correct up
The extracted watermark bits are then manipulated to to 11 bit errors anywhere while the below code length
recover the watermark information. A block diagram is 31 bits. There are actually 11 "data bits"' and 20
of the watermarking extraction process is shown in (31-11 =20) "check bits" within the code. The first 8
Figure 2. bits of the 11 data bits contains the bit sequence of
one watermark symbol, and the remaining 3 bits are
tl1acted I)strean
I( .1'1%'(1)7 dink Wa erIn
A r
fixed to '0'. Therefore, this BCH(3 1, 11) code can be
-x Iract in regarded as a BCH(31, 8) code with t 11, where t
refers to the error correcting capacity.
(I }1DC 1 = 1ilverse }Discr ete Coesine I'rans(l-vm
(I " = Inverse Quanltisation
M(l)1 =
Moli(II-compensated ilredition
= 312
3*. T ro 1
Turbo(3,J
VI) = Variahle- length Decoder
Figure 2 The block diagram of watermark The code rate of this turbo code is 1/3. The encoder
extraction system with MPEG-2 decoder structure of this turbo code and its component
convolutional code are shown in Figure 3 and Figure
4, respectively.
Authorized licensed use limited to: Moradabad Institute of Technology. Downloaded on December 20, 2008 at 01:31 from IEEE Xplore. Restrictions apply.
u =x state (so s1)
00
x|p
No RSC 1
~ ~ ~ ~ RSC 0
1 1
--------
10
Figure 7 Trellis diagram of the convolutional code (2,
1, 3)
4 Testing results
Figure 4 Encoder structure of the component
convolutional code In this section seven cases of noise were simulated
and tested. In order to compare the error correcting
3.1.3 Conv(2,1,3) capacity of these three error correcting codes, the
Watermark Correcting Rate (WCR) of each error
The convolutional code implemented has a code correcting code was measured under different Bit Error
rate of 1/2 and the constraint length is 3. The generator Rate (BER). BER is defined as:
is given by g= [I 1 1] (from left to right) and BER = (e I n) x 100%
where e is the number of bits in error in the
[10 1]. The encoder structure can be seen
extracted watermark bitstream before ECC decoding,
in Figure 5. and n is the total number of bits in the extracted
OLItput bit sequenices watermark bitstream before ECC decoding. WCR is
defined as:
WCR = (w / m) x 100%
[nput bit sequjence where w is the number of correct bits in the
extracted watermark data after ECC decoding, and m is
Figure 5 Encoder structure of (2, 1, 3) Convolutional the total number of bits in the extracted watermark data
code after ECC decoding. The tests were conducted under
the following seven cases of simulated noise:
Based on the encoder structure above, we can get
the trellis diagram as shown in Figure 7. Using the
binary sequence of watermark data "abcdefgh" as Case no. Description
input, the encoded watermark bitstream has 128 Case 1 Set the value of 1/4 part of Y components in e
I-frames to 0
(64 X 2 =128) bits, as shown in Figure 6. Case 2 Set the value of all Cr components in each
I-frames to 0
001101011100001110000101110011101111 Case 3 Set the value of all Cb and Cr components
010111001101010001011111101100110101 in each I-frames to 0
111110001000010111110101111101011111 Case 4 Set the value of 1/4 part of Y components and
01100100010100101100 all Cb and Cr components in I-frames to 0
Figure 6 (2, 1, 3) Convolutional code codeword Case 5 Set the value of all Y components in I-frame
sequence to half of the original value
From the above codeword, we find that the Case 6 Set the value of all Y, Cr components in
watermark data itself is not included in the encoded I-frames to half of the original value
Case 7 Set the value of all Y, Cb and Cr component
bitstream and is different from the codeword of in I-frames to half of the original value
BCH(31, 8).
Authorized licensed use limited to: Moradabad Institute of Technology. Downloaded on December 20, 2008 at 01:31 from IEEE Xplore. Restrictions apply.
According to the results in Figure 8 we find that WCR after turbo decoding remains at about 95% when
BCH (31,8) can achieve a higher error correcting the BER is around 15% or lower, and drops to around
capacity than Turbo (3,1) and Conv(2,1,3). When the 75% when BER reaches 20%. The poor performance is
BER is around 10% or lower, BCH(31, 8) can always partially due to the small size of the component
recover all the watermark information bits in errors and interleaver. For turbo code, in order to achieve the good
obtain the right watermark information. When the BER error correcting capacity, the size of the component
is around 15%, the WCR after BCH decoding still interleaver in a turbo codec needs to be very large.
remains above 95% and the extracted watermark However in this test, the size of the interleaver is set to
information is almost correct.The test results in Figure 8 be equal to the number of bits in the watermark data
show that the Turbo(3,1) can not obtain 100% correct before turbo encoding, which is only 64 when using the
information under seven cases simulated noise test. The plain test "abcdefgh" as watermark data.
ECC typ( BCH (31,8) Turbo (3,1) Conv (2,1,3)
BER (0) WCR (%) Extracted BER (%) WCR (%) Extracted BER (0) WCR (0) Extracted
Watermark Watermark Watermark
Data Data Data
Case 1 6.6484 100 abcdefgc 5.5556 95.125 abcdefgh 9.8485 100 abcdefgh
Case 2 7.6613 100 abcdefgc 8.5859 95.3125 abcdefgh 12.1212 100 abcdefgh
Case 3 14.5161 95.3125 abcdefgc 15.6566 95.3125 ab defgh 18.8394 71.8750 a'h¶em
Case 4 20.5645 76.5625 $B#,,af§" 21.2121 76.5625 A '&Eh 28.7879 56.2500 T Xm"141
Case 5 33.0968 66.6250 pt_U uiN KW 33.6019 49.8750 bsd5CeX 33.3507 59.8039 AdxqqI62
Case 6 41.4597 59.2812 -_yuN_KD 42.0673 50.3437 wb3b!KfiZ 41.8697 50.6434 tU0GolM¶
Case 7 49.6774 51.0938 yaOl-A 50.3981 50.8437 fEd1fl [uaZ 50.2162 51.8076 idR5 Ee
Figure 8 BER and WCR comparison of three types ECC under the seven cases of simulated noise
Authorized licensed use limited to: Moradabad Institute of Technology. Downloaded on December 20, 2008 at 01:31 from IEEE Xplore. Restrictions apply.