USING INVISIBLE WATERMARKS TO PROTECT VISIBLY WATERMARKED IMAGES

USING INVISIBLE WATERMARKS TO PROTECT VISIBLY WATERMARKED IMAGES

C.M.Chen

Outline

Introduction Framework Robust Invisible Watermarking Algorithm Experimental Results Future Work

Introduction

Several visible watermarking techniques have been proposed in the literature, and meanwhile, some problems with visible watermarks are also under investigation

1. watermark removal

2. unauthorized insertion

Introduction

In this paper, we propose using an invisible watermark in visibly watermarked images to overcome these problems

Visible watermarks are useful for protecting online images because they discourage unauthorized copying

Several visible watermarking techniques have been proposed in the literature

Introduction

Generally, visible watermarking requires that each watermark should be easily visible, unobtrusive, and hard to remove

Although we can use a number of methods to make the visible watermark difficult to remove, we have to admit that removal is not impossible

Introduction

some researchers have attempted to remove the embedded watermark by using image inpainting techniques

a visible watermark bearing a certain logo does not constitute a proof of ownership

some one can insert the logo of others within an image and claim that the resulting image comes from them

Introduction

the first is that a visible watermark must be difficult to remove

the second is that it must be able to withstand the impersonator problem

Most efforts in the literature focus on the first problem

However, to the best of our knowledge, there are few papers available for the second problem

Introduction

Mohanty et al. presented a dual watermarking technique which attempts to establish the owner’s right to the image and detect the intentional and unintentional tampering of the image

They claimed that if anybody tries to tamper the visible watermark intentionally, they can know the extent of tampering with the help of invisible watermark detection algorithm

Introduction

However, they obviously did not distinguish between the tampering of the visibly watermarked image and that of the embedded visible watermark

it is unable to tell whether those changes are targeted at the embedded visible watermark or the visibly watermarked image

Introduction

Wong and Memon used an invisible authentication watermark to ensure the identity of a visibly watermarked image

Any modification to the visible watermark would be reflected in a corresponding error in the fragile watermark

However, their scheme is too sensitive to be used in most practical applications

Some apparent disadvantages of using fragile watermarks for this purpose1. Common image processing such as compre

ssion, filtering, noise addition or geometric distortion can not hinder the embedded visible watermark from indicating ownership, but they can destroy the fragile watermark easily

Some apparent disadvantages of using fragile watermarks for this purpose2. When the owner’s visible watermark is

visually removed or tampered, or replaced by another unauthorized visible watermark, one can not identify the right owner to the image with the fragile watermarking scheme

Motivation

We need an invisible watermark that can enhance ownership protection of the visibly watermarked image

Ownership assertion watermarks are typically robust

So we need a robust instead of fragile invisible watermarking scheme in the dual watermarking system

The properties and associated requirements of such robust watermarks1)The watermark should be invisible and has n

o apparent interference with the visibly watermarked image

2)The watermark is desirously extracted without resorting to the visibly watermarked image

3)The watermark must be difficult to remove and can resist non-malicious changes such as image compression and malicious attacks such as image inpainting/replacement

The properties and associated requirements of such robust watermarks Visibly watermarked images are usually

compressed for online use So malicious attacks may be targeted at the

compressed-decompressed versions of images

The most challenge to such a watermarking technique is to make the invisible watermark robust against operations like compression and image inpainting

Robust invisible watermarking algorithm Since content-preserving image processing s

uch as compression has low-pass nature, it is reasonable to choose the low-pass component for watermark embedding

In this paper, we propose to embed the invisible watermark in the low-pass subband of a three-level wavelet decomposition image

Robust invisible watermarking algorithm Image inpainting/replacement will affect all th

e frequency components of any watermarking scheme based on a global transform, whereas the DWT-based scheme produces watermarks with local spatial support

The selection of three-level decomposition is to make the number of low-pass coefficients large enough for watermark embedding

Robust invisible watermarking algorithm

[7] toreferred becan embedding of steps detailed The

sequence watermarkof

index single theis ),...,2,1( and 1or 0 Here

sequencebinary thebe ][

, watermarklet the and image, atermarked visibly wthe

of tscoefficien pass-low thedenote ][Let

Kkm

mM

XX

k

k

k

Constructing invisible watermarks In this paper, we choose the watermark in the

form of the binary image of the embedded visible watermark so that the extracted logo can indicate the ownership without additional computing

To increase the security of the invisible watermark, the invisible watermark image is shuffled with some techniques like chaotic mapping before embedding (Here we use the Arnold’s cat map to transform the binary logo)

Constructing invisible watermarks After transformation, a binary block which foll

ows uniform distribution and has the same size as the binary logo is EX-ORed with the binary logo to create the encrypted watermark

The seed of the pseudo-random number generator is used as the private key

Constructing invisible watermarks In watermark detection, without the private ke

y, one can only obtain a meaningless binary sequence. However, legal users can recreate the binary logo by EX-ORing the extracted block with the pseudo-random block generated with the private key

Post-processing extracted invisible watermarks From experiments we notice that image oper

ations like compression and inpainting/replacement have different effects on the extracted binary image

The former produces random noise-like pixels across the entire extracted image whereas the latter only ruins parts of the image

Post-processing extracted invisible watermarks This is because compression could alter valu

es of some low-pass coefficients and thus cause parts of embedded information bits not to be extracted correctly

With higher compression ratio, more noise-like pixels would appear

Post-processing extracted invisible watermarks On the other side, inpainting/replacement, wh

ich is often a kind of spatial operation, only causes information bits embedded in targeted areas to be lost

To remove noise-like pixels resulting from compression, we use a median filter to post-process the extracted binary image

After such a non-linear filtering, most isolated noise-like pixels can be removed

Experimental results

Experimental results

Experimental results

Future research

Future efforts will focus on how to enhance robustness of the embedded invisible watermark, for example, to make the watermark against geometrical attacks which challenge all existing invisible watermarks

How to embed more watermark information is another major concern.

The research on dual watermarking system is very significant for the practical application of visible watermarks

The End

Inpainting

the technique of modifying an image in an undetectable form, is as ancient as art itself

The goals and applications of inpainting are numerous, from the restoration of damaged paintings and photographs to the removal/replacement of selected objects

Back