ENEE408G Lecture-9 Last Lecture Digital Watermarking and Fingerprinting for Digital Rights Protection of Multimedia URL: http://www.ece.umd.edu/class/enee408g/ Slides included here are based on Spring 2012 offering in the order of introduction, image, video, speech, and audio. Copyrighted 2002-2012. 2012 ENEE408G course was developed @ ECE Department, University of Maryland, College Park. Inquiries can be addressed to Profs. Ray Liu (kjrliu@isr.umd.edu) and Min Wu (minwu@eng.umd.edu). Audio synthesis: MIDI Digital Audio Coding/Compression Psychoacoustics properties used in perceptual audio coding MPEG-1 Audio coding Today: Digital Rights Management of Multimedia via Watermarking ENEE408G Capstone -- Multimedia Signal Processing ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [2] Demands on Info. Security and Protection Digital Watermarking/Data Hiding in Multimedia Intellectual property management for digital media Promising electronic marketplace for digital music and movies Advantages of digital: perfect reproduction, easy transmission, Napster controversy Conventional encryption alone still leaves many problems unsolved Protection from encryption vanishes once data is decrypted Still want establish ownership and restrict illegal re-distributions How to distinguish changes introduced by compression vs. malicious tampering? Bit-by-bit accuracy is not always desired authenticity criterion for MM What is Digital Watermarking? Examples: Picture in picture, words in words Silent message, invisible images Secondary information in perceptual digital media data The need of watermarking: robust vs. fragile Copyright protection: prove the ownership Fingerprinting: trace the source Copy protection: prevent illegal copying Data authentication: check authenticity of data Fragile or semi-fragile watermarking ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [3] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [4]
Example on Invisible and Robust Watermark Fragile Watermark Example: Document Authentication (a) (c) alter 10011010 (b) (d) (e) (f) after alteration Copyright Embed pre-determined d pattern or content features beforehand Verify hidden data s integrity to decide on authenticity (g) ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [5] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [6] General Framework of Data Hiding Basic Requirements for (Robust) Watermarking 101101 Hello, World 101101 Hello, World host media data to be hidden extracted data embed Player; or Forensic Analyzer play/ record/ extract marked media (w/ hidden data) compress process / attack test media ENEE408G Sl lides (created by R. Liu & M.W Wu 2002) Imperceptibility (perceptual transparency) Payload the amount of information that can be stored in a watermark Robustness Security Kerckhoff Principle The method used to encrypt the data is known to an unauthorized party and that the security must lie in the choice of a key. Blind and non-blind detection (aka Oblivious vs Non-oblivious) Blind detection ~ does not use the original unmarked copy ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [7] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [8]
search Talks 98-04) based on Res ated by M.Wu Data Embedding by Replacing LSBs search Talks 98-04) based on Res ated by M.Wu Data Embedding by Replacing LSBs (cont d) 31 Slides (crea ENEE63 Replace LSB with Pentagon s MSB 31 Slides (crea ENEE63 Replace 6 LSBs with Pentagon s 6 MSBs ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [10] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [11] by R. Liu & M.W Wu 2002) lides (created b ENEE408G Sl => See Lab Project 4 for details A Simple Audio Watermark in Time Domain Put message in the Least-Significant-Bits (LSBs) Encode a message into bits e.g., represent a character string into bits using ASCII code Embedder puts in LSBs of audio samples Repeat embedding the same bit in a few samples if needed Detector retrieves embedded bits from LSBs Perform majority voting if repeated embedding is used Repack bits into message Tradeoff between perceptual quality and robustness Compare the embedding in 1 st LSBs, 2 nd LSBs, Security Can they see/hear your message? Can other people make imperceptible change to alter your message? ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [12] by M.Wu & R. ENEE408G Slides (created => See Design Project 4 for details A More Robust Watermark in Transform Domain Embedder: use HAS & embed in perceptually significant freq. Audio File Key/Seed Segmentation (frame size L) Noise-like seq. Watermark DCT W(j) added in midfrequency region V (j) = V(j) + a(j) W(j) e.g., a(j) = 0.05 V(j) Detector: determine the existence of a specific wmk Subtract host signal, measure similarity (via correlation), & threshold it Audio File in question DCT V + _ V(j) Extract t mid-frequency coefficients V (j) watermark detector idct Detection result Watermarked audio file V W Original A specific Similarity il it measures: DCT audio file watermark <V V, W>, or < (V V)./ a, W >, or ENEE408G Capstone -- Multimedia Signal Processing correlation MM Digital coeff. Rights [13]
Discussions Digital Fingerprinting and Tracing Traitors by M.Wu & R. L lides (created b ENEE408G Sl Why use noise-like sequence as watermark? Imperceptibility p Confidentiality of the embedded data Robustness against jamming Imperceptibility Frequency domain embedding: can take advantage of known perceptual properties such as masking Can apply sophisticated HAS models to improve perceptual quality Robustness and security Use attacks to find weaknesses and improve designs Case study: SDMI public challenge (Fall 00) ted by M.Wu 2005) Leak of information poses serious threats to government operations and commercial markets e.g., pirated content or classified document studio The Lord of the Ring Promising countermeasure: robustly embed digital it fingerprintsi w1 w2 w3 Alice Bob Insert ID or fingerprint (often through robust watermarking) to identify each user Purpose: deter information leakage; digital rights management provide post-delivery protection complementary to encryption Carl Sell Challenge: imperceptibility, robustness, tracing capability ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [14] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [15] Fingerprinting Curves Embedded Fingerprinting for Multimedia 2005) 8G Slides (created by M.Wu Original Curve (captured by TabletPC) Fingerprinted Curve (100 control points) Detection Statistics Typical threshold is 3~6 for fl false alarm of f10-3 ~ 10-9 ted by M.Wu 2005) Embedded Fingerprinting Multi-user Attacks Traitor Tracing Customer s Multimedia ID: Alice Document Fingerprinted Copy Digital Distribute 101101 embed Fingerprint to Alice Alice Bob Fingerprinted doc for different users Unauthorized re-distribution Collusion Attack Colluded Copy (to remove fingerprints) Extract Identify Alice, 101110 Fingerprints Traitors Bob, Suspicious Copy Codebook ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [16] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [17]
Collusion-Resistant Fingerprinting of Maps Example of Anti-Collusion Fingerprint Code: Embed 16-bit Code for Detecting ti 3C Colluders Out of f20 2006) ted by M.Wu 8G Slides (crea MCP ENEE408 UM 5-User Averaging Attack 2-User Interleaving Attack... Also survive combination attacks of collusion + print + scan ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [18] 2005) 8G Slides (created by M.Wu User-1 ( -1,-1, -1, -1, 1, 1, 1, 1,, 1 ) Embed fingerprint via HVS-based spread spectrum embedding in block-dct domain Collude by Averaging ( -1, 1, 1, 1, 1, 1,, -1, 1, 1, 1 ) User-4 Extracted fingerprint code ( -1, 0, 0, 0, 1,, 0, 0, 0, 1, 1, 1 ) Uniquely Identify User 1 & 4 ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [19] ted by M.Wu 2005) 8G Slides (crea Case Study: Tracing Movie Screening Copies Potential civilian use for digital rights management (DRM) Copyright industry $500+ Billion business ~ 5% U.S. GDP Alleged Movie Pirate Arrested (23 January 2004) A real case of a successful deployment of 'traitor-tracing' mechanism in the digital realm Use invisible fingerprints to protect screener copies of pre-release movies Last Samurai w1 Carmine Caridi Russell friends Internet http://www.msnbc.msn.com/id/4037016/ Hollywood studio traced pirated version Summary Multimedia watermarking for rights management Reading Assignment F. Hartung and M. Kutter: Multimedia Watermarking Techniques, Proc. of the IEEE,,pp pp.1079-1107, July 1999. M. Wu and B. Liu, Multimedia Data Hiding, Chapter 10 on SDMI audio watermark challenge, preprint, 2002 (electronic handout). M. Wu, W. Trappe, Z. Wang, and K.J.R. Liu: "Collusion Resistant Fingerprinting for Multimedia", IEEE Signal Processing Magazine, Special Issue on Digital Rights Management, pp.15-27, March 2004. http://www.ece.umd.edu/~minwu/public p _paper/jnl/0403fpcollusionp _ IEEEfinal_ SPM.pdf This week s Lab session: Continue on audio project ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [20] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [21]
To Explore Further: Type-I Additive Embedding Slides (created b P ENEE408G S For Further Explorations: General Formulation of Two Types of Watermarking Add secondary signal in host media Representative: spread spectrum embedding Add a noise-like signal and detection via correlation Good tradeoff between imperceptibility p and robustness Limited capacity: host signal often appears as major interferer 10110100... data to be hidden modulation < X + noise, >= < + (X + noise), > X = X + marked copy X original source < X + noise - X, > = < + noise, > ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [23] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [24] Type-II Relationship Enforcement Embedding Type-II Relationship Enforcement (cont d) 8G Slides (crea Deterministically enforcing relationship Secondary information carried solely in watermarked signal feature value 2kQ (2k+1)Q (2k+2)Q (2k+3)Q odd-even mapping lookup table mapping 0 1 0 1 0 1 1 0 Representative: odd-even embedding No need to know host signal (no host interference) High capacity but limited robustness Robustness achieved by quantization or tolerance zone even 0 odd 1 E.g. enforcing black pixel# per block to odd/even to hide data in binary image General approach: Partition host signal space into sub-regions each region is labeled with 0 or 1 marked sig. is from a region close to orig. & labeled w/ the bit to hide Secondary info. carried solely in X difference (X -X) doesn t necessarily reflect the embedded data { b} data to 1 or 0 be hidden Advanced embedding: mapping X host sig. X = f( b ) marked copy Combining the two types with techniques suggested by info. theory ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [25] ENEE408G Capstone -- Multimedia Signal Processing MM Digital Rights [26]