WATERMARKING DAN

STEGANOGRAFI

CLO 2 : Memahami persepsi manusia

dalam hal audibilitas dan visibilitas yang
dimanfaatkan sebagian algoritma
watermarking/steganografi untuk
menyembunyikan informasi.
COURSE LEARNING OUTCOME
 CLO 1 : Mengetahui konsep dasar watermarking dan
steganografi, baik bidang keilmuan, perbedaan, tujuan, dan
manfaatnya. (10%)
 CLO 2 : Memahami persepsi manusia dalam hal audibilitas
dan visibilitas yang dimanfaatkan sebagian algoritma
watermarking/steganografi untuk menyembunyikan
informasi (20%)
 CLO 3 : Memahami metode watermarking/steganografi
berdasarkan domain tertentu menggunakan beberapa alternatif
transformasi dan proses penyisipan serta pengambilan
informasi kembali. (25%)
 CLO 4 : Mengaplikasikan proses watermarking/steganografi
dengan perancangan simulasi menggunakan Matlab (25%)
 CLO 5 : Mengaplikasikan serangan yang diterapkan pada
watermarking/steganografi untuk uji ketahanan watermark
(20%)
OUTLINE
 Karakteristik Telinga Manusia
 Psychoacoustic dan Modelnya
 Dynamic Frekuensi Masking
 Time Masking
 Persepsi Fasa
 Karakteristik Mata Manusia
 Psychovisual
 Parameter Kinerja Objektif (PEAQ, ODG, SNR,
PSNR)
 Parameter Kinerja Kualitas Informasi Yang
DIekstraksi Sesuai Jenis Watermark (BER, CER,
SSIM, NC, DR)
Ear
Psychoacoustics
 Principles of the human perception of sound
 MPEG compression algorithm uses model of
human hearing to remove data (perceptual
coding algorithm)
 Frequency range is about 20 Hz to 20 kHz, most
sensitive at 2 to 4 KHz.
 Dynamic range (quietest to loudest) is about 96
dB
 Normal voice range is about 500 Hz to 2 kHz
 Low frequencies -> vowels, bass; High ->
consonants

5 of 17
Human Hearing Sensitivity
 Experiment: Put a person in a quiet room. Raise
level of 1 kHz tone until just barely audible. Vary
the frequency, plot:

6 of 17
Human Frequency Masking
 Experiment: Play 1 kHz tone (masking tone) at
fixed level (60 dB). Play test tone at a different
level (e.g., 1.1 kHz), and raise level until just
distinguishable.
 Vary the frequency of the test tone and plot the
threshold when it becomes audible

7 of 17
Frequency Masking

CPSC 538a MPEG Audio Tutorial January 12,

2004 8 of 17
Temporal Masking
 If we hear a loud sound, then it stops, it takes a little while until we
can hear a soft tone nearby.
 Experiment: Play 1 kHz masking tone at 60 dB, plus a test tone at
1.1 kHz at 40 dB. Test tone can't be heard (it's masked). Stop
masking tone, then stop test tone after a short delay.
 Adjust delay time to the shortest time when test tone can be heard
(e.g., 5 ms).
 Repeat with different level of the test tone and plot:

9 of 17
Temporal Masking
Combination

11 of 17
Phase Sensitivity

12 of 17
Phase Sensitivity
If x is audio signal, then we change x to –x. We
will hear the same audio without any distortion.
-x  x with phase modification 180o

Example :
We have x1=sin(2*pi*f1*t) and
x2=sin(2*pi*f1*t+pi/3) and x2=cos(2*pi*f1*t+pi/5)
x1, x2, x3 will have the same audio human
perception
In other meaning, human can not absolutely
detect the audio delay accurately
13 of 17
Eye
Anatomy of the Eye
 Formasi Citra
 Cahaya dipancarkan dari sumber cahaya
 Cahaya dipantulkan oleh object.
 Cahaya yg dipantulkan ditangkap oleh mata atau camera
 Cahaya adalah radiasi elektromagnetis yang menstimulir respons
visual, dan diekspresikan sebagai distribusi energi spectral L(l), dimana
l adalah panjang gelombang antara 350nm – 780 nm.

16
 Light
UMCP ENEE631 Slides (created by M.Wu © 2004)

Figure is from slides

at Gonzalez/ Woods
DIP book website
(Chapter 2)

 Light is an electromagnetic wave

 with wavelength of 350nm to 780nm stimulating human visual response
 Expressed as spectral energy distribution I(l)
 The range of light intensity levels that human visual system can adapt is
huge: ~ on 10 orders of magnitude (1010) but not simultaneously
 Brightness adaptation: small intensity range to discriminate
simultaneously
 Photoreceptors
Rods – sensitive to dim light
(night vision)
- found mostly in periphery
- why can see a dim star better if look a
few degrees to one side

Cones – sensitive to color

(daylight vision)
- concentrated in fovea
- provides fine detail
light
Cones  RGB Color
 s-cones absorb short wavelength light best,
with peak response at 450 nm (blue)
 L-cones absorb long wavelength light best,
with peak response at 580 nm (red)
 i-cones absorb intermediate wavelengths L-cones
best, with peak response at 540 nm (green) i-cones
 Light at any wavelength in the visual
s-cones
spectrum from 400 to 700 nm will excite
these 3 types of cones to a degree
depending on the intensity at each Spectral response of cones in typical human eye
wavelength.
 Our perception of which color we are
seeing (color sensation) is determined by
how much S, i and L resonse occurs to
light of a particular intensity distribution.
Rule: To get the overall response of each type of
cone, multiply the intensity of the light at each
wavelength by the response of the cone at that
wavelength and then add together all of the
products for all of the wavenumbers in the
intensity distribution
CONVENTIONAL OBJECTIVE
QUALITY MEASURES
• Signal to Noise Ratio (SNR)  Mostly used
• Segmental SNR (segSNR)
• Cepstral Distortion Measure (CD)
• Log-Likelihood Ratio Measure (LLR)
• Itakura Saito Distortion Measure (IS)
• Log Area Ratio Measure (LAR)
• Weighted Spectral Slope Measure (WSS)
• Total Harmonic Distortion Measure (THD)
 SNR

sw(n) = watermarked audio

s0(n) = host audio
ITU-R BS.1387-1 METHOD FOR
OBJECTIVE MEASUREMENT OF
PERCEIVED AUDIO QUALITY
• SUBJECTIVE DIFFERENT GRADE
• OBJECTIVE DIFFERENT GRADE  FOR AUDIO : PEAQ
(PERCEPTUAL EVALUATION FOR AUDIO QUALITY )
SUBJECTIVE DIFFERENT GRADE
(SDG) / Mean opinion score (MOS)

SDG = GradeSignal Under Test – GradeReference

Signal
SDG ITU-R BS. 1387 five Imperceptibility
Scale grade impairment
(ITU-R scale / MOS
BS.562)
0 5 Imperceptible
-1 4 Perceptible But Not
Annoying
-2 3 Slightly Annoying
-3 2 Annoying
-4 1 Very Annoying
OBJECTIVE DIFFERENT
GRADE (ODG) / PEAQ (1)
OBJECTIVE DIFFERENT
GRADE (ODG) / PEAQ (2)
OBJECTIVE DIFFERENT
GRADE (ODG) / PEAQ (3)
• AUDIO : PERCEPTUAL EVALUATION FOR AUDIO QUALITY
(PEAQ)
• SPEECH : PERCEPTUAL EVALUATION FOR SPEECH QUALITY
(PESQ)
• VIDEO : PERCEPTUAL EVALUATION FOR VIDEO QUALITY
(PEVQ)
 ODG scale

SDG ODG Scale / PEAQ Imperceptibility

Scale
0 0 Imperceptible
-1 -1 Perceptible But Not
Annoying
-2 -2 Slightly Annoying
-3 -3 Annoying
-4 -4 Very Annoying
 SIMILAR WITH PEAQ (PREVIOUS
VERSION OF ODG FOR AUDIO)
 DISTURBANCE INDEX (DI)

 NOISE TO MASK RATIO (NMR)

 PERCEPTUAL AUDIO QUALITY

MEASURE (PAQM)

 PERCEPTUAL EVALUATION (PERCEVAL)

 PERCEPTUAL OBJECTIVE MEASURE

(POM)
CONVENTIONAL METHOD
• SIGNAL TO NOISE RATIO FOR DIFFERENCE
(SNR) MEASUREMENT, BUT
• TOTAL HARMONIC STILL USED IN AUDIO
DISTORTION (THD) QUALITY MEASUREMENT,
ESP. SNR
Robustness measurement

• BINARY DATA (BIT) : BER

• DETECTION RATE : DR
• TEXT : CER
• WORD : WER
• IMAGE : SSIM, PSNR, NC, SDG
• AUDIO : ODG/PEAQ, SNR, SDG
 BER, DR, NC

BER :

DR :

NC (Image) : Rumusnya salah,

harusnya ada j di
sigmanya
SSIM (STRUCTURAL SIMILARITY)

Taux dan
tau y
tidak
dikuadra
tkan 
cek
PSNR (PEAK SNR) AND MSE
Soal Latihan/Kuis
1. Jelaskan definisi filter psikoakustik !
2. Apa maksudnya dynamic masking dalam domain frekuensi?
3. Jelaskan temporal masking pada audio !
4. Bagaimanakah visibilitas manusia terhadap warna? Ruang warna apakah yang sensitifitas
mata manusia paling rendah? Jelaskan !
5. Bagaimanakah representasi fasa yang dirasakan indera manusia?
6. Bagaimana teknik penyisipan data pada audio dengan memanfaatkan dynamic masking? Jelaskan
usul teknik penyisipannya dan ekstraksinya sebanyak minimal 3 teknik!
7. Bagaimana teknik penyisipan data pada audio dengan memanfaatkan temporal masking? Bagaimana
kira-kira secara umum proses ekstraksinya?
8. Buatlah script untuk menghitung PSNR dan SNR! Serta tunjukkan dimana bedanya!
9. Buatlah script untuk menghitung BER dan SSIM!
10. Buatlah script untuk menghitung NC!
11. Buatlah script untuk menggunakan ODG!
12. Jika x dan y adalah sinyal audio mono 8 bit (0-255), dimana x=[1 2 10 30 100] dan y=[12 3 11 28 90],
a. Hitung SNR antara x dan y !
b. Hitung MSE antara x dan y !
c. Hitung PSNR antara x dan y !
d. Apa bedanya SNR dan PSNR dari perhitungan diatas ? Jelaskan!
e. Hitung NC x dan y !
13. Jika x dan y adalah citra 2x2 8 bit grayscale, x=[50 150;200 170] dan y=[45 155;200 180],
a. Hitung SNR antara x dan y !
b. Hitung MSE antara x dan y !
c. Hitung PSNR antara x dan y !
d. Hitung NC x dan y !
e. Hitung SSIM x dan y !