SEGMENTATION BASED IMAGE AUTHENTICATION FOR IMAGE FORGERY CLASSIFICATION AND LOCALIZATION

34

SEGMENTATION BASED IMAGE AUTHENTICATION

FOR IMAGE FORGERY CLASSIFICATION AND

LOCALIZATION

MANIMEGALAI.A

M.E (CSE)

Jay Sriram Group of

Institution

[email protected]

A.GOKILAVANI

Assistant Professor

Jay Sriram Group of

Institution

[email protected]

DR.RAJALAKSHMI

HOD(CSE)

Jay Sriram Group of

Institution

[email protected]

ABSTRACT

The proposed scheme in this paper addresses the problem of recognizing image forgery in image authentication. Consequently, in recent years several hashing methods have utilized machine learning to improve the hashing quality by learning a group of Patches functions.However, in the learning of patches functions, they either are sensitive to the data distributions or ignore the correlations of patches functions. In this project, we propose a new forgery detection method, namely, image forgery classification and localization with local models, for image authentication. Propose method is developed for detecting image forgery including removal, insertion, and replacement of objects and for locating the forged area. The local models include position and texture information of object regions in the image. Secret keys are introduced in segmentation as feature extraction and patches construction. While being robust against content-preserving image processing, the patches is sensitive to malicious tampering and, therefore, applicable to image authentication. The patch of a test image is compared with that of a reference image. When the patches distance is greater than a threshold and less than, the received image is judged as a fake. By decomposing the patches, the type of image forgery and location of forged areas can be determined. Probability of collision between patches of different images approaches zero. Image authentication tecniques are presented to show effectiveness of the method.

KEYWORDS

Image forgery detection,Image authentication,Localization,Segmentation.

I. INTRODUCTION

Copy –move forgery in an image contains atleast

couple of regions whose contents are identical.

Normally the aim of a forger is to either cover the

truth or to enhance the visual effect of the image.

CMFD is one of the most important and popular

digital forensic techniques currently used .In the

literature, there are mainly two classes of CMFD

algorithms. One is based on block-wise detection and

other on keypoint extraction. They both try to detect

the CMF through describing local patches of one

image. Keypoint extraction method is preferred

mostly because it requires less computational

resource than block-based ones. In CMFD scheme,

test image is first segmented into non-overlapped

patches, then matching process between patches is

taking place in two stages

.

the suspicious pairs of patches that may contain

copy-move forgery regions, and we roughly estimate

an affine transform matrix. In the second stage, an

Expectation-Maximization-based algorithm is

designed to refine the estimated matrix and to

confirm the existence of copy move forgery.

Eventhough proposed CMFD achieves better

performance than other methods, there usually a large

35

consuming.Furthermore,some of these detected

patches may be just false alarm containing not any

CMF regions. Our proposed work focus on efficient

and automatic techniques is desired to identify and

verify the contents of digital images. Image

authentication is such a promising technique to

automatically identify whether a query image is a

different one, or a fabrication, or a simple copy of an

anchor image. Image hashing is a technique that

extracts a short sequence from the image to represent

its contents, and therefore can be used for image

authentication. If the image is maliciously modified,

the patches must be changed significantly.

Meanwhile, unlike patches functions in cryptography

such as MD5 and SHA-1 that are extremely sensitive

to slight changes in the input data, the image Patches

should be robust against normal image processing.

The objective is to provide a reasonably short image

patches with good performance and services of our

proposal system are discussed.

II. CMFD SYSTEM AND IMAGE

SEGMENTATION

CMFD is proposed to detect the copy-move forgery

in an image, mainly by extracting the keypoints for

comparison. Usually ,block based methods need a

large amount of time to detect an image. So its

important to decrease the number of patches for

comparing. In this regard keypoint-based methods

are faster and more favorable than block-based ones,

because the number of keypoints are smaller than that

of the divided blocks. However, on the other hand,

key point-based method also has the following two

problems. Firstly, the key points lying spatially close

to each other should not be compared because they

may be naturally similar. The determination of the

shortest distance between two comparable key points

is tricky. Most prior arts empirically select this

threshold but neglect its relationship with the image

size and content. Secondly, it is uneasy to accurately

localize and distinguish the copying source region

and the pasting target region, because, unlike the

overlapping blocks, the key points are often not

concentrated together. To deal with this problem this

method was further improved as clustering –based

CMFD scheme significantly raise the accuracy of

localization of CMF regions. First have to segment

the image into number of non-overlapped patches.

Then CMFD is performed by matching these patches

,as long as the pasting target and copying source

regions are not in the same patch. In our CMFD

scheme, after segmenting the image, we perform the

first stage of affine estimation. During this stage we

first extract the keypoints from the whole image and

construct a k-d tree. Then the KNN search is

performed in each region for each keypoint to find a

possible correspondence. One region is recorded if it

has certain proportion of keypoints matched with

another one. Finally estimate the affine relationship

36

is the input to the second stage of matching process,

where we iteratively refine the matrix via a

probability model based on EM algorithm.

Image Segmentation

In order to separate the copying source

region from the pasting target region, the image

should be segmented into small patches, each of

which is semantically independent to the others.

In our implementation, each image is empirically segmented into no less than 100 patches, and thus, a CMF region may be in two or more patches. In consequence the useful information for CMFD is reduced in each patch. However, to obtain a convincing detection result we need not a large number of key points (sometimes four is enough). Furthermore, because the CMF region exists in many patches, we meanwhile have more than one chance to find the tampering operation.

Matching Process of Our Proposed CMFD System

The steps involved in this stage are,

 Key point Extraction and Description – Patches generation from detected objects.

 Matching Between Patches - Define the distance between two key points.

 Affine Transform Estimation - Estimate the relationship between source region and pasting target region.

 CMF Determination Based on Probability - Exploit all the pixels in the matched patches to find out a more accurate estimation.

 Obtaining the New Correspondences of the Pixels - Five neighboring pixels are considered and find the new matching pixel.

 Iterative Re-Estimation of the Transform Matrix - compute the correlation coefficients between the transformed image and the original test image. Generally speaking, matching process consist of the following three steps.

1) Obtaining the matched points. 2) Calculating the transform matrix.

3) Repeating the above two steps until a convergence condition is satisfied.

III. PROPOSED WORK

37

In general, a good image Patches should be reasonably short, robust to ordinary image manipulations, and sensitive to tampering. It should also be unique in the sense that different images have significantly different Patches values, and secure so that any unauthorized party cannot break the key and coin the Patches. To meet all the requirements simultaneously, especially perceptual robustness and sensitivity to tampering, is a challenging task.

The objective is to provide a reasonably short image Patches with good performance, i.e., being perceptually robust while capable of detecting and locating content forgery.

The services provided by the proposed image authentication system include:

 Identify a query image as a similar image, or a tampered image, or a different image, w.r.t. a Saliency image.

 Evaluate similarity of two images by distance between them.

 Identify and locate three types of tampered area, i.e., added area, removed area, changed area.

 Estimate the percentage of tampered area. When an image is sent to a user, a possible solution to prove the authenticity is to generate a

Patches value and send it securely to the user. The Patches value is a compact string – an abstract of the content. A user can re-generate a patches value from the received image, and compare it with the original Patches value. If they match, the content is considered as authentic. In order to allow incidental distortion, the Patches value must possess some robustness.

EDGE DETECTION MECHANISMS

Our proposed workincludes researching on the embedding algorithm robust to geometric distortions and improving the precision in locating

the altered areas by implement via any digital multimedia networking application for verify the content of image transmission over RGB features. So this kind of implementation is desired to find features that better represent the image contents so as to enhance the patches’ sensitivity to small area tampering while maintaining short patches length and good robustness against normal image processing like object edge detection mechanisms and also include tracer routing to detect the content modified hacker system which is use full to reduce the hacking possibilities. So without knowledge of this method, hacker information may be acknowledged to the sender once hacker receives the packet for content or object modifications.

Routing technique

Tracer routing is to find out the unauthorized router access i.e the system modify the content of the image and forward to the destination in a routing process. This is verifying by getting packet processing time from each and every router in the routing process by a destination. Then destination find out the timing differences with all routers if any timing is differed then it will be consider as unauthorized ip.

IV. IMPLEMENTATION

1. IMAGE SEGMENTATION AND FEATURE EXTRACTION

 A new method to construct robust and secure image patches using Global Features, which is based on luminance and chrominance characteristics of the image.

 The image is first processed. The pre-processing steps include re-sizing using bi-linear interpolation.

38

 First segments the preprocessed image into semantically independent patches prior to key point extraction.

 Segment the global features - As the

luminance component of an object in a preprocessed image contains most structural and textural information of an object i.e. evaluate brightness ratio of an image and

chrominance represents RGB extraction

from an object.

 Segment the local features – Coarseness

extraction which is the parameter can be estimated heuristically from the contrast of textures in an object.

 The final patches sequence is obtained by pseudo-randomly permuting the binary sequence from the global features and local features.

2. KEYPOINTS BASED PATCHES GENERATION

 The global and object local vectors are concatenated to form an intermediate patches.

 Which is then pseudo-randomly scrambled based on a secret key to produce the final patch sequence.

 Here use advanced encryption algorithm to encrypt the patch sequence with respect to secret keys.

3. IMAGE TRANSFER AND VERIFICATION When an image is sent to a receiver, a possible solution to prove the authenticity is to generate a patch value, encrypt it and send it securely to the receiver. The patch value is a compact string – an abstract of the object in an image. A receiver can re-generate a patch value from the received image i.e. follow the same segmentation patch generation procedure at the sender side, then

decrypt the received patches successfully and compare it with the received im2age patch values. If they match, the content is considered as authentic. In order to allow incidental distortion, the patch value must possess some robustness.

4. PATCHES DISTANCE MATCHING

 We use a patches distance between patches of an image pair as a metric to judge similarity/ dissimilarity of the two images.

 The patch sequence of a received image to be tested with the decrypted hash sequence under similarity ratio if difference is above the threshold then it has been maliciously tampered or legitimate image.

 The method can be used to locate tampered areas and tell the nature of tampering, e.g., replacement of objects or abnormal

5.FORGERY DETECTION AND LOCATE

 After dissimilarity comparisons of patches, the both patches are applied in the received image.

 By seeing the images with original patches and received side generated patches, the system concludes and locates the forgery area in the received image i.e. by shading the mismatched part.

6.IP TRACER ROUTING

Tracer routing enables the destination to trace the whole routing process and find out the unauthorized router access in a routing process. Destination verifying by getting and differentiate the packet timing from each and every router in the routing process.

V. CONCLUSION

In this work, we propose efficient robust

segmentation based image authentication system to

39

using IP tracer routing .This proposal achieves a very high security level. The proposed image authentication system could serve as a building block in many applications such as copyright protection, image retrieval and video signature.This proposal has better performance of discriminating high quality images from malicious attacks than some existing schemes.

REFERENCES

1. M.A.Fischler and R.C.Bolles, ”Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography,” Commun. ACM, vol. 24,no. 6, pp. 381-395,Jun. 1981.

2. A.J.Fridich, B.D.Soukal, and A.J.Lukas, “Detection of copy-move forgery in digital images,” in Proc. Digit. Forensic Res. Workshop, 2003 3. R.Hartley and A.Zissserman, multiple view geometry in computer vision,2nd ed. Newyork, NY, USA :Cambridge Univ.press,2004.

4. D.G.Lowe,”Distinctive image features from scale invariant keypoints”, Int. J. Comput. Vis., vol. 60, no. 2, pp. 91-110, Nov .2004.

5. H.Bay, A.Ess, T. Tuytelaars, and L. Van Gool,”SURF:speeded up robust features,” Comput.

Vis. I mage Understand.,vol.110, no. 3, pp. 346-359, Jun 2008.

6. H.Huang, W.Guo, and Y.Zhang, “Detection of copy-move forgery in digital images using SIFT algorithm,” in Proc. Pacific-Asia Workshop Comput. Intell. Ind. Appl. (PACIIA), vol. 2, Dec. 2008, pp.889-892.

7. S.Bayram, H.T. Sencar, and N.Memon, “An efficient and robust method for detecting copy-move forgery,” in Proc.IEEE Int.Conf.Acoust., Speech Signal Process.(ICASSP), Washington, DC, USA, Apr.2009, pp.1053-1056.

8. M.Akhin and V.Itsykson, “Clone Dtection: why, what and how?” in Proc. 6th Central Eastern Eur. Softw. Eng . Conf. (CEE-SECR)2010, pp. 36-42.

9. E.Ardizzone, A.Bruno, and G.Mazzola,” copy-move forgery detection via texture description,” in Proc. 2nd ACM Workshop Multimedia Forensics, Secur. Intell., Newyork, NY, USA, 2010,pp. 59-64