You are on page 1of 19

International Conference on Electronics, Communications and Information Technology

(ICECIT) – 2021
14–16 September 2021, Khulna, Bangladesh

Paper ID: 383

Analogy among Different Conventional


Watermarking Transforms and Noise
Removing Filters against Various Attacks
Presented By:
Sejuti Zaman
Contents :
• Introduction
-digital watermarking
-general approach
• Objectives
• Methodology
• Result Analysis
-performance parameters
- performance analysis
• Conclusion

2
Introduction

1. Digital Image Watermarking


-Process of hiding a message within an ordinary
image
-extraction of this message at its destination
-Imperceptible changes of the ordinary image
(host image)

3
Introduction
2. General Approach
1.Spatial domain
Watermark
2.Frequency domain

Watermarked Image
Cover Image
Embedding
Procedure

Watermarking Key

Figure 1 Basic digital watermarking embedding process

4
Introduction
2. General Approach (cont)
1.Blind procedure
2.Non-blind procedure

Watermarked image Extracted watermark

Extracting
procedure

Watermarking key Distortion of image for


different types of attack

Figure 2 Basic digital watermarking extraction process

5
Objectives
1. Basic objectives

• To analyze conventional transforms like FFT, DCT and


DWT
• To find out more reliable method to do watermark
• To analyze the performance parameters like MSE and
PSNR
• To find out an appropriate filter for corresponding noise
attack

6
Methodology
Original RGB to
DCT Vectorization
Image Gray

+
Watermark RGB to Time
Vectorization *
Image Gray Domain

Filtered Noisy Watermarked


Extraction IDCT
Image Image Image

PSNR & PSNR &


MSE MSE

Fig 3 Block diagram of invisible watermarking based on


sinusoidal transform (DCT transform)
7
Methodology
Original RGB to
FFT Vectorization
Image Gray

+
Watermark RGB to Time
Vectorization *
Image Gray Domain

Filtered Noisy Watermarked


Extraction IFFT
Image Image Image

PSNR & PSNR &


MSE MSE

Fig 4 Block diagram of invisible watermarking based on


sinusoidal transform (FFT transform)
8
Methodology
Original RGB to CA,CH, CH,
DWT Vectorization
Image Gray CV,CD CV,CD

Watermark RGB to Time


Vectorization * +
Image Gray Domain


Merge
Filtered Noisy Watermarked CA,CH,CV,CD
Extraction
Image Image Image
IDWT
PSNR & PSNR &
MSE MSE

Fig 5 Block diagram of invisible watermarking based on non-


sinusoidal transform (DWT Transform)
9
Methodology
1.Sample Images

Lena Peppers Boats Cameraman

Jet-plane Lake MRI Ultrasound

10
Methodology
2.Simulation

Cover image

Embedded watermarked image

Watermark image

11
Methodology
Restored original
2.Simulation output

Noise attack Denoise image

Extract code
12
Methodology
2.Simulation output

Restored MRI image


MRI image

Watermarked
image

Code image Extracted code


Figure 6 Medical images

13
Result Analysis
3. Performance Parameters
• PSNR values for different types of noise:

SNRdb = 10 * log10

• MSE values for extracted original image:

MSE=

14
Result Analysis
4. Performance analysis
Table 1 MSE AND PSNR(DB) OF ORIGINAL VS
WATERMARKED IMAGES OF DIFFERENT
DCT TRANSFORMS
FFT DWT
Images
MSE PSNR MSE PSNR MSE PSNR
Lena 3.62 60.2292 4.67 88.2879 584.30 21.1767
Peppers 3.73 59.4503 4.58 87.5194 584.12 21.4242
Boats 3.65 59.1975 4.55 87.2473 584.35 21.0820

Cameraman 3.60 60.4785 4.65 88.5297 584.16 21.3322

Jetplane 3.59 59.8453 4.59 87.6703 584.04 21.6945


Lake 3.60 59.9521 4.70 88.0031 584.01 21.7240
MRI 1.78 63.6767 1.11 94.7432 584.28 21.2080
Heart 1.72 59.5117 1.80 82.6561 583.89 21.4039
Ultrasound 1.69 62.7487 1.46 92.4049 584.42 19.2228
15
Result Analysis
4. Performance analysis
Table 2 MSE AND PSNR(DB) OF ORIGINAL VS
RECONSTRUCTED IMAGES OF DIFFERENT
DCT TRANSFORMS
FFT DWT
Images
MSE PSNR MSE PSNR MSE PSNR
Lena 0.2446 30.6861 3.16 29.84 40.61 56.86

Peppers 1.9232 23.3523 15.57 23.26 47.83 46.84

Boats 2.3597 22.4640 19.27 22.16 50.52 43.69

Cameraman 0.4453 29.7056 4.16 29.00 44.35 56.73

Jetplane 1.1011 25.7744 16.27 23.04 48.39 45.44

Lake 0.8619 26.8382 7.26 26.54 42.46 50.46

MRI 0.0926 36.5257 0.51 38.18 11.27 69.20

Heart 2.5302 21.4885 5.38 27.91 37.72 58.70

Ultrasound 0.0658 28.3690 3.64 29.50 36.38 58.53

16
Result Analysis
4. Performance analysis
Table 2 PSNR (DB) VALUES OF NOISY VS FILTERED
IMAGES (PEEPERS) BY DWT
Noisy Wiener Gaussian Median Average
Noise image filter filter filter filter
Gaussian 18.76 45.7602 44.7313 45.6244 45.2584

Poisson 22.16 46.9509 45.2941 46.8770 46.4043

Speckle 21.42 46.7676 45.1175 46.6964 46.3638

Salt & 17.29 42.4422 45.4791 46.4624 44.8242


Pepper

17
Conclusion
In this work, three watermarking methods are evaluated
against an image dataset containing images from multiple
domains. The robustness of the algorithms are studied with
different noise attacks and showed promising results. DWT
transform performs better and more reliable and against
various attacks, median filter is working well. In the future,
deep learning based convolutional neural network approach
can be explored to improve the performance of
watermarking. The study will be helpful in image
watermarking baseline experiments.

18
19

You might also like