MULTIPIXEL VISUAL CRYPTOGRAPHY FOR COLOR IMAGES INFORMATION TECHNOLOGY BY A.Anil (006-09-5001) R.

Gaurav Singh (006-08-5012)

UNDER THE GUIDANCE OF P.Jhansi Rani

DEPARTMENT OF INFORMATION TECHNOLOGY VASAVI COLLEGE OF ENGINEERING HYDERABAD

CERTIFICATE

This is to certify that the project report entitled Multipixel Visual Cryptography for Color Images is a bonafide work done by the students A.Anil and R.Gaurav Singh in partial fulfillment of the requirement for B.E 4/4 semister-2(IT) of Osmania University, Hyderabad during 2011-2012.

Head of the department

(Mrs.N.Vasantha)

Project Coordinator (Mr.K.Aravind Kumar)

Project Guide (P.Jhansi Rani)

ACKNOWLEDGEMENTS

We are extremely grateful to our project guide for excellent guidance for the course of the project .We are grateful to our guide for providing all the possible help and constructive suggestions to sort out the problems that arose during the work.

We are grateful to the management and the principal of Vasavi College of Engineering for providing quality education with excellent infrastructure.

We are grateful to Mrs.N.Vasantha, H.O.D., Dept. of I.T., for her encouragement and guidance .We express our sincere thanks to her genial mentoring.

Multi-pixel Visual Cryptography for color images

Abstract: Visual Cryptography is a special encryption technique to hide information in images in such a way that it can be decrypted by the human vision if the correct key image is used. Visual Cryptography uses two transparent images. One image contains random pixels and the other image contains the secret information. It is impossible to retrieve the secret information from one of the images. Both transparent images/layers are required to reveal the information. Using secret sharing concepts, the encryption procedure encrypts a secret image into the so-called shares which are noise-like secure images which can be transmitted or distributed over an unsecure communication channel. Using the properties of the human visual system to force the recognition of a secret message from overlapping shares, the secret image is decrypted without additional computations and any knowledge of cryptography. Any visual secret information (pictures, text, etc) is considered as image and encryption is performed using simple algorithm to generate n copies of shares depending on type of access structure schemes. The simplest access structure is the 2 out of 2 scheme where the secret image is encrypted into 2 shares and both needed for a successful decryption. These shares are random dots without revealing the secret information. Visual cryptographic solutions operate on binary or binaries inputs. Therefore, natural (continuous-tone) images must be first converted into halftone images by using the density of the net dots to simulate the original gray or color levels in the target binary representation. Here, a halftone image is made up of a series of dots rather than a continuous tone. These dots can be different sizes, different colors, and sometimes even different shapes. Larger dots are used to represent darker, denser areas of the image, while smaller dots are used for lighter areas. Then, the halftone version of the input image is used instead of the original secret image to produce the shares. The decrypted image is obtained by stacking the shares together. Existing System: Images are manipulated by the attackers in the network. Confidential Images has no means to be secured when they are transmitted over the network.

In the traditional visual cryptography schemes, only one piece of image was encoded during encryption. And each of the shares that are generated was shown as a disorganized image. Due to the capacity of encrypted information, these schemes have few applications and because of the disorganized image share, they are easily suspected by hacker. The generated shares are not meaningful.

Proposed System:

Visual Cryptography for color images to generate two meaningful shares. Some filters are proposed for better visual quality of recovered image. The input image is decomposed into three channels of Cyan, Magenta and Yellow by using equation. Error Diffusion produces halftone of much higher quality than other halftone. Halftone is applied on each monochrome image. It is applied on each halftone channel to generate shares. There will be two shares for each halftone. A multi-pixel non-expanded scheme for color images introduced which can encode more than one pixel for each run resulting same size of shares as secret image. A new simple watermarking algorithm is proposed to generate meaningful shares. Decryption is achieved by stacking the shares. In case of black pixel, overlaying two rows of M1 results in four black bits, and reveals the information, where as for the white pixel, stacking the two rows of M0 results in two black and two white bits, and thus introduces noise.

Table of Contents

1. Motivation 2. Introduction 3. Secret Sharing Schemes 4. Experiment Result 5. Improvement Methods 6. Conclusion 7. Future Work 8. Reference

1 Motivation
The traditional secret sharing scheme can share secret among a group of members according to some formulas, such as polynomial or a set of equations. While surfing the Internet, we found a novel way to sharing a secret by embedding information into several images. To retrieve the concealed information, all we need to do is stake some of these images and then we will see the secret appear before our eyes. This puzzle-like structure attracts our attention, and we also eager to know the specific implementation methods. Therefore, we want to study and implement the visual secret sharing schemes.

2 Introductions 2.1What is secret sharing?

Secret sharing was invented by both Adi Shamir and George Blakley independently in 1979. Secret sharing is a way to distributing secret among a group of members such that each member owns a share of the secret. Only some specific combinations of shares can reconstruct the secret. Individual shares reveal nothing about the secret. Visual Secret Sharing is an encryption technique that doesnt need complex calculations in order to decrypt a message. The ciphertext and the key consist of transparencies (or images). When properly stacked these transparencies (or stacked by image editing package), one can reveal the plaintext.

2.2 What we have done

Survey and study some famous visual secret sharing schemes (VSS) and audio secret sharing schemes (ASS). Implementation of VSS using c++. Including (2, 2), (2, n), (n, n), additional schemes and weighted schemes. Propose an idea about video secret sharing. Implementation and analysis are mentioned, too. Discuss the two major drawbacks of VSS (Expansion and Contrast). Some ongoing solutions are addressed.

3 Secret Sharing Schemes

3.1 Visual Secret Sharing Schemes (VSS)

Visual Secret Sharing (VSS) is proposed by Naor and Shamir [1]. Different from other secret sharing schemes, VSS target on being easily decoded by the human visual system without the knowledge of cryptography and cryptographic computations, that is, negligible computing power compared to those is required for reconstructing the original information. This section introducing the fundamental secret sharing scheme we surveyed and utilized in our project implemented. Before the scheme is discussed, we first define the (k, n) VSS as follow: (1) n secret shares are constructed according to the original image, and the shares are also called shadows. (2) k is the threshold number of shares required to reconstruct the original image. That is, the original image can be decoded with any combination of k shadows (2-1); yet no information can be captured with any combination of k shadows, where k < k (2-2). Along with the definition, each pixel in the original image will be mapped to m sub-pixels in each of the n shadows which the mapping involves a random factor. By superimposing (or-ing) k *m corresponding sub-pixels of k shadows, we can get m sub-pixels. Whenever k >= k, given b and w are the numbers of black and white sub-pixels out of the m sub-pixels gained by the superimposition, respectively, b/w should be different depending on whether the original pixel is black or white so that the original image can be recognized by the contrast, achieving definition (2-1). And to ensure definition (2-2), whenever k < k shadows are superimposed, b/w should be the same no matter black or white the original pixel is. To decide the m sub-pixels of each of the n shadows, m by n 0-1 matrices are defined for the case where the original pixel is black and the case where the original pixel is white. The m by n scalars in the matrices correspond to the m sub-pixels of the n shadows, and their 0/1 values mean the corresponding sub-pixels are white/black. The matrices are called the Code Blocks. When utilizing the Code Blocks, the corresponding matrix is selected with the m columns randomly permuted, and then the n candidate shadows (the n rows) are randomly distributed to the n shadows. Following is an example of (2,2) schemes: Pixel p = 0.5 first shadow second shadow superimposed

p = 0.5

p = 0.5

p = 0.5

c1=

0 1

1 0

1
c0 =

0 0

Figure 1. An example of (2,2) schemes

3.2 Audio Secret Sharing Schemes (ASS)

Audio secret sharing was first proposed in 1998 by Desmedt, Hou, and Quisquater [2]. To embed a message into a stanza of cover sounds is the basic idea of the scheme. Further, every share of the stanza sounds normal and message-free. The general representation of it is (k, n) ASS, just like the VSS scheme, and k represents the threshold number while n stands for the number of shares. Several methods are introduced in the following paragraphs. Other than the DHQ ASS by Desmedt and Hou and Quisquater, two methods proposed by our fellow Taiwanese are introduced. One is Cadence ASS by and , and the other one is the Time Division ASS by , , and [3]. DHQ ASS The phase difference of sound waves, say, constructive interference and destructive interference, is utilized. For instance, while the two are in the same phase, the composite

amplitude of the two sound waves increases, and so does the volume. On the contrary, the volume decreases while the phase difference is . The transformation is illustrated below.

Figure 2. An example of ASS with triangle cover sound

In the Fig.2.[2], a sequence of plaintext message (0 0 1 1) is transformed into share1 and share2 using the triangle wave cover sound. In the DHQ scheme, two shares are constructed with the employment of a wave of cover sound. Therefore, [log2n] cover sound waves are used while producing n shares. What is more, the scheme is only suitable for (2, n) ASS. Cadence ASS The scheme could be mapped to VSS. Instead of the volume in the DHQ, the number of cadences within a time unit is used to represent the information of a bit in this scheme. If the number of cadences is larger than the threshold, a 1 is embedded, and vice versa. In (t, n) ASS scheme, a variable r is added to control how many pulses compose a time unit. Considering a (2, 2)8, tow kinds of rhythms are produced by a cover sound wave. These two rhythms, such as Ra and Rb, would generate two voices at the last pulse.

Figure 3. A (2, 2) schemes in ASS

The matrix [Ra Ra, Rb Rb] is used for embedding a 0, and [Ra Rb, Rb Ra] is for a 1. According to this principle, when playing the two shares shown below at the same time, a message (1 0 0 1) is revealed.

Figure 4. An example of share1 and share2

Following is the mapping between VSS and ASS.

Figure 5. Mapping between VSS and ASS

3.3 Video Secret Sharing Schemes

As we introduced in the previous paragraphs, the basic operation of a visual secret sharing scheme is overlapping images in space. This inspire us a new idea: how about overlapping images in time domain? The persistence of vision( ) in human vision system brings out the same effect as stacking while fast showing two images in order. A video secret sharing scheme is proposed by utilizing this human vision property. After the video secret sharing process, we expect there is a sequence of frames that tell nothing individually but display a secret while playing successive frames speedily. Therefore, an eyedropper cant get information by knowing frames that are discontinuous. We denote the ith frame which is processed to be f i, ith reconstructed frame frei. (fi is analogous to shares in VSS.) To implement the concept, we apply VSS for each frame in original video sequence, such that fi + fi+1 = frei. The experimental result shows that the secret can be reconstructed by our scheme, and the pixels in individual frames seem to be randomly distributed so that they reveal nothing about the original secret. However, there are some weaknesses in this scheme. First, video file size is quite large after this process. Whats worse, compression dont work since the random distribution nature of pixels in these frames. Second, a small amount of information leakage occurs due to the dependency between fi and fi+2.

4 Experiment Result

1. Visual secret sharing with (2, 2) scheme:

Figure.6: shadow 1

Figure.7: shadow 2

Figure.8: reconstructed image

The secret information is hidden into two shadows in Figure.6 and Figure.7. Put the two shadows together, the secret message could be recognized from the reconstructed image, which is in Figure.8. In the reconstructed image, a pixel is divided into four blocks. Two of the four blocks are randomly chosen to be black if the pixel should be white while all of the four blocks are colored black if the pixel should be black.

Figure.9: shadow 1

Figure.10: shadow 2

Figure.11: reconstructed image

The secret information is hidden into two shadows in Figure.9 and Figure.10. What is the difference between these two shadows and Figure.6/Figure.7 is that these two shadows contain some information on its own, not only a picture of mosaic. Put the two shadows together, the secret message could be recognized from the reconstructed image, which is in Figure.11. In the shadows, a pixel is divided into four blocks. Two of the four blocks are randomly chosen to be black if the pixel should be white while three of the four blocks are randomly chosen to be black if the pixel should be black. However, after overlapping the two shadows, three of the four blocks are black if the pixel should be white while all of the four blocks are colored black if the pixel should be black. 2. Visual secret sharing with (2, 4) scheme:

Figure.12: shadow 1

Figure.13 shadow 2

Figure.14: reconstructed image 1

Figure.15: reconstructed image 2

The secret information is hidden into four shadows. Put arbitrary two of the four shadows, say, Figure.12 and Figure.13, together, the secret message could be recognized from the reconstructed image, which is in Figure.14. If three or four shadows are put together, the color of reconstructed image would be darker and the secret information in it would become clearer. Figure.15 is the reconstructed image after overlapping all of the four shadows.

3. Visual secret sharing with (3, 3) scheme:

Figure.16: shadow 1

Figure.17: shadow 2

Figure.18: shadow 3

Figure.19: reconstructed image

The secret information is hidden into three shadows. Put all of the three shadows, Figure.16 and Figure.17 and Figure.18, together, the secret message could be recognized from the reconstructed image, which is in Figure.19. Attention. There would be no information leakage while putting only two of the three shadows together.

4. Visual secret sharing with (2, 2) scheme with a hierarchical structure:

Figure.20: shadow 1

Figure.21: shadow 2

Figure.22: reconstructed image 1 Figure.23: shadow 3

Figure.24: reconstructed image 2 Figure.25: shadow 4

Figure.26: reconstructed 3

There is not only one secret image imbedded in the shadows. A hierarchical structure makes the secrets appear in a special time pattern. The first secret information, Figure.22, would be seen after putting the first two shadows, Figure.20 and Figure.21, together. Adding the third shadow, Figure.23, to Figure.22, the second secret information would be released, which is in Figure.24. In the end, the last shadow in Figure.25 is put on the second secret information image and then the final secret information would be seen in the last reconstructed image, Figure.26.

5 Improvement Methods

5.1 Reduce Image Size in VSS

In VSS, a pixel substituted by m sub pixels results in an increased size of shares. That is a great issue to figure out the minimum m for a VSS scheme. Moreover, many researches based on probability are trying to reduce the shares size by the price of lessened resolution. In [6], Yang proposes a probabilistic method such that the sizes of the original image and shadows are the same, that is, non-expansible. And it can be easily implemented on a basis of conventional VSS scheme. In [7], he proposes another concept of trading the shadow size for the contrast. In this paper, they study size-adjustable VSS schemes such that one can choose appropriate shadow size and the recovered image contrast for practical use.

5.2 Increase Image Contrast in VSS

Another drawback of visual secret sharing schemes is much loss of contrast in the reconstructed image. Since some more operations other than simple addition are feasible in the real world application, many new schemes are proposed in order to get higher contrast quality. For instance, operations such as XOR or inversing can be done on tokens and smart cards. The paper [8] presented by Duong Quang Viet and Kaoru Kurosawa shows a new paradigm in which the original image is almost perfectly reconstructed. A very simple reversing operation is used, exchanging black and white. The concept is illustrated below: Pixel inv(T) inv(T) inv(T) + inv(T) inv(inv(T) + inv(T))

p = 0.25

p = 0.25

p = 0.25

p = 0.25

p=1

Figure 27. An illustration of inversing scheme T = s1 + s2 and T = s1 + s2 s1, s2 are two shares of a general (2,2) VSS schemes. s1 and s2 are another run of (2,2) shares. inv() denotes inversing operation. We can straightforwardly understand that only 1/2 white pixels will be reconstructed to a mix-color block. The others are perfectly white or perfectly black. After showing this (k, n) scheme that white pixels are almost perfectly reconstructed in addition to the perfect reconstruction of black pixels, they next show how to convert a perfect black (k, n)(with reversing) into a perfect white (k, n)-VCS with reversing. Thirdly, they show a perfect black VCS for any monotone access structure and to colored and grey level images.

6 Conclusion

In this final project, we survey and implement some visual secret sharing schemes on simple black and white images. The feature of VSS is indeed attractive. However, we also experience personally some dramatic weakness of VSS while coding the program. Fortunately, many promising new schemes are continuously proposed in order to solve the disadvantages of VSS / ASS. To sum up, following is a list of advantages and disadvantages: Advantages Interesting easily decode(using human eyes or ears) simple decode operations such as addition, XOR and inversing can be done on ability-restricted machines Disadvantages For VSS, position accuracy is required; for ASS, time accuracy. image contrast decreases (VSS) image expansion (VSS)

7 Future Work

Visual / audio cryptography is indeed an attractive field which is worth studying. In the future, we want to survey VSS schemes of gray scale and color images, and extend the capability of our program such that suitable for any kinds of input images. Besides, we can try to figure out some refinement methods for current weaknesses in VSS and ASS.

8 Reference

[1] M. Naor and A. Shamir, Visual cryptography, Advances in CryptographyEUROCRYPT94, Perugia, Italy, pp. 1-12, May 1994. [2] Y. Desmedt, S. Hou and J. Quisquater, Audio and optical cryptography, in Advances in Cryptology-Asiacrypt'98, Springer-Verlag LNCS, pp. 392-404. [4] Chen-Chi Lin, Chi-Sung Laih, and Ching-Nung Yang, "New audio secret sharing schemes with time division technique," Journal of Information Science and Engineering, Vol.19 No.4, pp. 605-614 [5] G. Ateniese, C. Blundo, A. De Santis, D. R. Stinson, Extended capabilities for visual cryptography Theoretical Computer Science, Vol. 250 , Pages: 143 161 [6] Ching-Nung Yang, New visual secret sharing schemes using probabilistic method Pattern Recognition Letters, Volume 25, Pages: 481 494, Year of Publication: 2004 [7] Ching-Nung YANG and Tse-Shih CHEN, Size-Adjustable Visual Secret Sharing Schemes, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences 2005 E88-A(9):2471-2474; doi:10.1093/ietfec/e88-a.9.2471 [8] Duong Quang Viet and Kaoru Kurosawa, Almost Ideal Contrast Visual Cryptography with Reversing, CT-RSA 2004, LNCS 2964, Springer, pp.353-365 (