ANALYSIS OF IMAGE HASHING SCHEME BASED ON SECURE

PERCEPTUAL SIGNAL
SAHANA.M.S
ECE Department, MIT, S.R.Patna, Belwadi,Mandya, 571438, India

VEENA.S.K
ECE Department, MIT, S.R.Patna, Belwadi,Mandya, 571438, India

ABSTRACT: Perceptual image hashing is a process to estimate the distortions between the
images and maps an image to short binary string as on an image appearance in the visual domain.
This paper proposes full perceptual hashing systems that use a quantization module followed by a
cryptographic-compression module. This theoretical analysis is based on a study of the behaviour
of the extracted features that are modelled by Gaussian noise and then introduce a proposed
perceptual hashing scheme based on this theoretical analysis. At the end, several experiments are
conducted to validate the approach, by Gaussian noise and JPEG compression.

KEYWORDS: Image hashing, Image authentication, Quantization, Perceptual hashing, Feature

extraction, Robust hashing.

INTRODUCTION

Rapid development in multimedia and network technologies, the amount of information that is
conveyed, broadcast via digital devices has grown exponentially. Simultaneously un-authorized
use has reached a significant level. To ensure confidentiality, multimedia data should stay
unintelligible without a decryption key. This is achieved mainly through encryption secret key or
public key.

In one of the earliest approaches, Schneider et al.(1996) use intensity histograms of image blocks
for authentication. The verification process involves computing the Euclidean distance between
the histogram of the original and the candidate image to be verified. The sum of all such distances
over the image is used as a measure of image authenticity. This approach requires storage of public
key encrypted histograms which can be considerably large.

J. Fridrich et al.(2000) propose a robust hash based on preserving selected discrete cosine
transform (DCT) coefficients. Their method is based on the observation that large changes to low
frequency DCT coefficients of the image are needed to change the appearance of the image
significantly. To make the procedure dependent on a key, the DCT modes are replaced with DC-
free random smooth patterns generated from a secret key.

Kailasnathan et al.(2001) extract a signature that is based on intensity statistics such as mean,
variance and other higher order moments of image blocks. While simple in concept, their scheme
has drawbacks that it is easy to modify an image without altering its histogram (e.g. permuting
pixels within a block). This makes the scheme less secure.
Mihcak and venkatesan (2001) develop image hashing algorithm by using an iterative approach to
binarize the DC sub band (lowest resolution wavelet coefficients) of a 3-level Haar wavelet
decomposition of the image. The key observation in their work is that the significant geometric
features of an image are preserved under small perturbations to the image.

X.C. Guo et al.(2007) propose a perceptual hashing scheme based on a combination of the discrete
wavelet transform (DWT) and the Radon transform. Taking the advantages of the frequency
localization property of DWT and the shift/ rotation invariant property of the Radon transform, the
algorithm can effectively detect malicious local changes, while also being robust against content-
preserving modifications.

S. Xiang et al.(2007) propose a histogram-based perceptual image hashing function using the
resistance of two statistical features: image histogram in shape and mean value. However, the
fragility to malicious attacks is a drawback. An improvement of this method is proposed by Y. Ou
et al.(2009), where the authors propose an improved histogram- based image hashing scheme
using a K-means algorithm, which obtains a better fragility result than S. Xiang et al.(2011).

The common aspect between all of the above mentioned schemes is that they do not really take the
crypto-compression stage into account. When the crypto compression stage is missed, security
properties are threatened.

The aim of this paper is to develop a theoretical analysis of full perceptual hashing systems that
use a quantization module followed by a crypto-compression module. Section 2 presents the
proposed perceptual hashing method based on a theoretical analysis of the quantization problem in
perceptual hashing systems. In section 3, several experimental results are presented. Conclusions
are drawn in Section 4.

PROPOSED SYSTEM

This section propose perceptual hashing scheme that is robust to the quantization stage. Based on a
theoretical study of the quantization problem in perceptual hashing systems, I propose to add new
modules to the standard perceptual hashing system presented in Fig. 1. The block diagram of the
hash generation module is shown in Fig 1. Various steps involved in the hash generation process
are as follows:

1. Let the input image be represented by I of dimension N x M pixels. Image I is split in to

non-overlapping blocks of dimension q x p pixels. This gives a total of (N/q) x (M/p)
blocks. Each block is represented by , where i=1. . . (N/q) x (M/p)
2. Let represent the gray value of a pixel at spatial location in the block
, where k=1, . . . ,q x p. Let the mean of each block be represented by , where i is the
block index. Each is calculated as follows:

(1)

All of the computed continuous means present features extracted from the transformed
image in the feature extraction stage. Thus, they should be quantized during the

606
 quantization stage to form the quantized intermediate perceptual hash vector with a
specific quantization step Q. The uniform quantization technique is used in our scheme.
Let be the element of the quantized intermediate perceptual hash vector of a specific
index i. is calculated as follows:

(2)

Fig.1. Proposed perceptual hashing scheme robust to the quantization stage.

3. The sender determines the information about the desired robustness to the additive
Gaussian noise, . Based on this information, the analysis module gives the appropriate
randomly selected taking into account the desired robustness.
4. The quantized intermediate perceptual hash vector is compressed and encrypted by the
cryptographic hash function SHA-1.Consequently, the obtained final perceptual hash is
256 bits in size.

RESULTS AND DISCUSSION

In the experiments of the proposed perceptual hashing scheme, for each given quantization step
size, we tested different image block sizes against different additive Gaussian noise levels. Fig. 2
shows an example of an original image and their noisy versions.

607
 (a) (b) (c) (d) (e)

(f) (g) (h) (i)

Fig. 2. Original image and noisy versions with different additive Gaussian noise parameterized
with different standard deviations : (a)original image, (b) =1 (c) = 4 (d) =6 (e) =10 (f) =13
(g) =20 (h) =31 (i) = 45

Similarity evaluation
An evaluation of the perceptual similarity between the original and the modified versions is based
on the perceptual aspect provided by the human visual system (HVS). Table 1 gives the SSIM and
PSNR values for noisy images.

Table 1. SSIM and PSNR values for noisy images obtained by applying different standard
deviation values of the additive Gaussian noise.
Standard SSIM PSNR (dB) Image quality Perceptual hash
deviation
1 0.987 45.98 Very similar
4 0.943 34.24 Similar
6 0.839 29.24 Similar
10 0.817 28.96 Similar
13 0.740 24.06 Different
20 0.688 21.34 Different
31 0.416 17.82 Very different
45 0.322 16.29 Very different

Robustness Evolution
Gaussian noise
Consider the originals image shown in Fig. 3(a). In Fig. 3(b), as lightly Gaussian noisy version of
the image . While in Fig. 3(c), the additive Gaussian noise of a standard deviation . For a good
perceptual hashing robustness and discrimination level, I must have: H(I)= and H(I) .

608
3.2.2 JPEG compression
Fig. 4(a) and(b)shows the JPEG compressed images of the original image. For a robust perceptual
image hashing system, we should have: H(I)= and H(I) .

(a) (b) (c)

Fig. 3. Original Gold hill image and its Gaussian noisy versions (a)Original Gold hill image I,
(b)Gaussian noisy =6 Gold hill image and (c)Gaussian noisy =14 Gold hill image

(a) (b)
Fig.4. JPEG compressed images of the Gold hill image in Fig. 4(a)JPEG compressed (QF = 70%)
Gold hill image and (b)JPEG compressed (QF = 10%) Gold hill image

Comparison with other methods

Our proposed approach allows us to generate a secure perceptual hash of 256bits and ensures the
robustness without wasting the storage space. Fig 5 compares my approach with others methods.

(a)

609
 (b)

Fig.5(a) Precision-recall graph (b) Comparision with other methods

CONCLUSION

This paper has addressed the theoretical aspects of the quantization stage in image hashing. We
have proposed a new perceptual hashing method that takes the quantization stage into account and
we have tested the perceptual scheme by several experiments to demonstrate the effectiveness of
the proposed model.

ACKNOWLEDGMENT

I would like to graciously thank Dr. B. G. Naresh Kumar, Principal, MIT, Mysore, Karnataka .I
finally like to thank colleagues involved at the MIT , who has provided me with the best
knowledge about the project and for their invaluable support and guidance.

REFERENCES

J. Fridrich and M. Goljan, Robust hash functions for digital watermarking, Proc. IEEE Int.
Conf. on Information Technology: Coding and Computing, pp. 178-183, Mar. 2000.
X.C. Guo, D. Hatzinakos, Content based image hashing via wavelet and Radon transform, in:
Proceedings of the Multimedia 8th Pacific Rim Conference on Advances in Multimedia
Information Processing, PCM07,Springer-Verlag, Berlin, Heidelberg, 2007, pp. 755764.
C. Kailasanathan and R. S. Naini, Image authentication surviving acceptable modifications
using statistical measures and k-mean segmentation, IEEE- EURASIP Work. Nonlinear Sig. and
Image Processing, vol. 1, June 2001.
K. Mihcak and R. Venkatesan, New iterative geometric techniques for robust image hashing,
Proc. ACM Workshop on Security and Privacy in Digital Rights Management, pp. 13-21, Nov.
2001.
Y. Ou, C. Sur, K. H. Rhee, An improved histogram-based image hashing scheme using k-means
segmentation, in: The Fourth Joint Work shop on Information Security(JWIS2009),Kaohsiung,
Taiwan, August2009.

610
 M. Schneider and S. F. Chang, A robust content based digital signature for image
authentication,Proc. IEEE Conf. on Image Processing, vol. 3, pp. 227-230, Sept. 1996.
S. Xiang, H.-J. Kim, J. Huang, Histogram-based image hashing scheme robust against
geometric deformations, in: Proceedings of the 9th Workshop on Multimedia & Security, ACM,
New York, NY, USA, 2007, pp. 121128.
S. Xiang, H.-J. Kim, Histogram-based image hashing for searching content-preserving copies,
Transactions on Data Hiding and Multi- media Security 6 (2011) 83108.

611