Digital Watermarking. Introduction Relation to Cryptography –Cryptography is Reversibility (no evidence) Established –Watermarking (1990s) Non-reversible.

DigitalWatermarking

Introduction

• Relation to Cryptography– Cryptography is

• Reversibility (no evidence)• Established

– Watermarking (1990s)• Non-reversible (noise)

– Information Hiding• Covert communication channel (steganography)

Digital Watermarking Media

• Video

• Audio

• Images– Our discussion will focus on this.

WatermarkingAlgorithm

WatermarkedImage

OriginalImage

WatermarkBlock Diagram

of image watermarking

Applications

• Copyright– The objective is to permanently and unalterably mark the image

so that the credit or assignment is beyond dispute.

• Digital Rights– A file may only be used by users with a license that matches the

watermarked signature.

• Information Hiding– Foil counterfeiters

• Revision History– Tamper detection

• Meta-tagging– Store keywords, descriptions, time along with images.

Criteria

• Main Criteria– Capacity– Payload – Computational Complexity– Transparency– Robustness

Require optimum relationshipRequire optimum relationship

Capacity

• The ability to detect watermarks with a low probability of error as the number of watermarks in a single image increases.

Payload

• The amount of information that can be legitimately stored within a data stream– Dependent on host medium– JPEG example

Computational Complexity

• Difficulty in process of watermark extraction– Realtime?

Transparency

• Transparency refers to the perceptual quality of the data being protected.– Watermark should be invisible over all image

types as well as local image characteristics.

• Need to consider perceptually insignificant portion of host image for insertion for maximum transparency

Robustness

• Resistance to attacks on the watermark– Attack – an operation performed on the image that

compromises the watermark– Active, Passive, Collusion, Forgery– Blind vs. Nonblind

• Use of non-robust watermarks– eg. tamper detection

Approaches and Implementation

• Two Types of Encoding– Spatial watermarking (spatial domain)– Spectral watermarking (frequency-domain)

• Many types due to variety of transforms• Adjustments made in frequency domain• More robust

Spatial-Domain Implementation

• Low-level Encoding

• Use of Image Analysis Operations– eg. Edge Detection/Color Separation

• Cons– Easily Attacked (Cropping)

Frequency-Domain Implementation

• Algorithm– Decomposition of image– Addition of Watermark

• Possibly encoded/encrypted

– Re-composition of Image

Frequency-Domain Implementation (Discrete Cosine Transform)

• Discrete Cosine Transform (DCT)– Used in today’s standard JPEG compression

• Relation to DFT• Compression explained by previous groups

– Image divided into non-overlapping blocks– Each block is DC transformed– Block coefficients are quantized through a special

algorithm

• Not ideal for human visual system

Frequency-Domain Implementation (Wavelet Transform)

• Wavelet Transform– Based on Short Time Fourier Transform

(STFT)– Becoming more common in compression

techniques• Better model of Human Visual System than DCT

Examples of Wavelets

Frequency-Domain Implementation(Common Wavelet Transform Algorithm - Decomposition)

Filter Bank Decomposition (10 Bands)

Frequency-Domain Implementation (Wavelet Transform Algorithm - Overview)

Watermarked Image

EncodedWatermark

Frequency-Domain Implementation (Cortex Transform)

• Cortex Transform– Recent– Mimics human visual system

• Corresponds to known structure of human eye

– Has its own disadvantages• Computational complexity – requires much more

data!

Other Issues

• Just Noticeable Difference (JND)– Threshold based on Human Visual System

• Adjustment in Frequency• Adjustments in Intensity

– Important impact on transparency

• Spatial adjustment of Frequency-Domain Watermark

Spread Spectrum

• Used to fulfill transparency criterion• The watermark in is based on spread spectrum

communications– Delivers narrowband data through a noisy channel, by

modulating each data symbol with a wideband (but very low amplitude) signal.

– The data is a single bit – a yes or no decision on whether the given watermark is present.

– The channel is the image data itself– The wideband signal is the watermark.

Color Images

• Scheme nearly identical to grayscale– R/G/B channels

• Each color plane treated as a separate image

– Luminance/Chrominance channels• Luminance = intensity• Chrominance = color

Resources• ftp://skynet.ecn.purdue.edu/pub/dist/delp/watermark-proceedings/paper.pdf• http://www.cosy.sbg.ac.at/~pmeerw/Watermarking/• http://www.cosy.sbg.ac.at/~pmeerw/Watermarking/MasterThesis/• http://www.eso.org/projects/esomidas/doc/user/98NOV/volb/node308.html• http://www.jjtc.com/Steganography/• http://www.mathworks.com/matlabcentral/files/3508/digital%20watermarking.pdf• Mihcak, Mehmet Kivanc. “Information Hiding Codes and Their Applications to Images

and Audio”, PhD Thesis. 2002.