Page 1
2006 2 20ポスター発表会 年 月 日 1/13
Two Channel Digital Watermarking for Musicbased on Exponential Time-spread Echo Kernel
Nguyen Thi Huong Lien04M35670 ([email protected] )
廣田研究室知能システム科学専攻・東京工業大学
楽曲に対する指数型時間拡散エコーカーネルに基づく2チャネル電子透かし法
審査員廣田薫教授 高橋宏教授 山村雅幸教授 室伏俊明助教授 高玉圭樹講師
発表時間: 14.00~15.30
Page 2
2006 2 20ポスター発表会 年 月 日 2/14
Digital Watermarking: Motivation & Background
watermark
EmbeddingExtraction/
yes/no
Detection
Digital Watermarking
Easy copy
Copyright Problems Imperceptibility Robustness
Capacity
Digital Watermarking: Trade-off
Page 3
2006 2 20ポスター発表会 年 月 日 3/14
Digital Watermarking: Attacks
Filter
Ambience
Conversion
Compression
Modulation
Time stretchClipping
t
magnitude
t
magnitude
Type 1: Affect magnitude of samples
Type 2: Affect time sequences(For ex.: clip a portion)
Page 4
2006 2 20ポスター発表会 年 月 日 4/14
Proposed 2 Channel Digital Watermarking
Embedding Process2channelseparation
Segment-tation
Segment-tation
LSB
EchoHiding
Encoding
Re-segment
Re-segment
2channelunification
originalsignal
watermarkedsignal
watermark
watermarklength
watermarkbits
Page 5
2006 2 20ポスター発表会 年 月 日 5/14
Least Significant Bit Channel (1/2)
16 bit quantization
watermark bit
watermark bit
1
0th 15th 9th
1 0 1 0 0 1
1 0 1 00 1
0
LSB
LSB (H. Yeh, J. Kuo, 1999): Robust to attacks of type 2
0
1
Original signal
0 t
Page 6
2006 2 20ポスター発表会 年 月 日 6/14
Least Significant Bit Channel (2/2)
LSB for Proposed 2 Channel DW
watermark
t
magnitude
1 segment = watermark lengthOriginal signal
Embed watermark into each segment
Page 7
2006 2 20ポスター発表会 年 月 日 7/14
Echo Hiding Channel (1/2)
originalsignal
echo signal
1d
(b) 1 kernel
2( )2
lx
K
originalsignal
echo signal
(c) 0 kernel0d
2( )2
lx
K
2i
EH (D. Gruhl, 1996):Robust to attacks of type 1
=
input output
original signal echo kernel echoed signal
(a) embedding process
original signal
echo
2( )lh
Parameters of echo kernelseffect sound quality
Page 8
2006 2 20ポスター発表会 年 月 日 8/14
Echo Hiding Channel (2/2)
EH for 2 Channel DW watermark
t
magnitude
1 0 0 1 0
1 kernel 0 kernel 1 kernel0 kernel 0 kernel
Original signal
Page 9
2006 2 20ポスター発表会 年 月 日 9/14
Exponential Time-spread Echo Kernel (1/2)
Single Echo MethodRoom Reverberation
i
originalsignal
echoesp
1KpK
Exponential time-spreadecho kernel
based on roomreverberation
delay
Page 10
2006 2 20ポスター発表会 年 月 日 10/14
Original music
Embeddingprocess Watermarked music
Experiment results: 2 Channel DW and ETEK (1/3)
Page 11
2006 2 20ポスター発表会 年 月 日 11/14
WatermarkExtraction
process
LSB channel EH channel
Can confirm the copyright
Experiment results: 2 Channel DW and ETEK (2/3)
Page 12
2006 2 20ポスター発表会 年 月 日 12/14
Type of attack Proposedsystem
LSB EH
equalization 72% 45.91% 72%
chorus 71.55% 44.12% 71.55%
cut freq.>15kHz 68.01% 44.43% 68.01%
reverse 62% 55.63% 62%
delay & reverse 57.42% 57.42% 50%
Robustness against attacks
4.57 3.86 5: Imperceptible
4: perceptible but not annoying
3: slightly annoying
2: annoying
1: very annoying
• Average results of sound quality from 7 young people• SONY, MDR-CD900ST Headphone (studio standard)
Experiment results: 2 Channel DW and ETEK (3/3)
Accuracy Rate
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10
Music clip
Perc
enta
ge (%
)
従来法提案法
Conventionalmethod
Proposedmethod
Single echo
ETEK
Page 13
2006 2 20ポスター発表会 年 月 日 13/14
Conclusions
Proposal: 2 channel digital watermarking system based on the LSB
method and EH Exponential Time-spread Echo Kernel for EH
Results: Can confirm copyright information Robustness to various attacks Natural sound quality
Page 14
2006 2 20ポスター発表会 年 月 日 14/14
Data used in experiments
Song (44.1 kHz, 16 bit) Genre
1. 有難う 1 Pop
2. 有難う 2 Pop
3. 明日は Pop
4. Go back Rock
5. お前の歌 Rock
6. Obata World
7. Path of the wind Ambient
8. Some other time House
9. 絶えない Gospel
10. 椿姫 Classic
List of music and genre used in experiments
Want to listen to the watermarked music?