technical overview of iwdw 2015 - 國立臺灣大學ipr/mmsec2015/data/lecture/technical...technical...

40
Technical Overview of IWDW 2015 Prof. Ja-Ling Wu Dept. of CSIE & GINM National Taiwan University

Upload: truongdung

Post on 24-Jun-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

Technical Overview of IWDW 2015

Prof. Ja-Ling Wu

Dept. of CSIE & GINM

National Taiwan University

• IWDW stands for International Workshop on Digital Forensics and Watermarking

• IDWD 2015 is the 14th workshop of IWDW series and is held at Tokyo University of Science, the Library Hall, Katsushika Campus.

• Date: 7 to 10, October, 2015

Program Schedule

• Oct. 7 : Registration/Reception

• Oct. 8:

Keynote 1: Advances of Video Forensics (Prof.

Alessandro Piva, U. of Florence, Italy)

A large part of the digital forensics activities have been devoted to analysis of still Images.

• Video Forensics is still an emerging research field, because

1. Creating fake images is much easier than

creating tampered videos

2. Images are usually available in a few possible

formats, while videos can be encoded with

many different coders,

3. Videos usually undergo a stronger

compression as compared with Images, making

the forensic analysis more difficult.

• In this invited talk, Prof. Alessandro introduced the most useful traces that can be used for video processing detection.

• Due to the ubiquitous diffusion of compressed videos, most of Prof. Alessandro’s talk will focus on traces relying on specific properties of the compression process.

• Ref: An overview on Video Forensics, APSIPA Trans. On Signal and Information Processing, August 2012.

• Oral Session 1: Forensics (I) 10:40 -12 :00 • There are 4 papers presented in this session: 1. Image Noise and Digital Image Forensics (France) 2. Camera Source Identification with Limited Labeled Training Set (China) 3. Detecting Video Forgery by Estimating Extrinsic Camera Parameters (China) 4. Discriminating between Computer-generated Facial Images and Natural Ones Using Smoothness Property and Local Entropy (Japan)

• The first paper offers an overall review of digital image noises, from their causes and models to the degradations they suffer along the image acquisition pipeline.

• The authors show that by the end of the pipeline, the noise may have widely different characteristics compared to the raw image, and consider the consequences in forensic and counter-forensic imagery.

• The second paper investigated the problem of model-based camera source identification with limited labeled training samples.

• The authors proposed an Ensemble projection (EP) method to deal with this challenge. Experimental results illustrate that the proposed EP method achieves a notable higher average accuracy than previous algorithms when labeled training samples is limited.

• In the third paper a geometric technique was proposed to detect tampering in video sequences.

• In this approach, extrinsic camera parameters, which describe positions and orientations of camera, are estimated from different regions in video frames.

• A statistical distribution model is then developed to characterize these parameters in tampering-free video and provides evidences of video forgery finally.

• Advances in technology have made it possible for computers to generate realistic multimedia contents that are very difficult to distinguish from non-computer generated contents. The authors of the 4th paper have created a method for identifying computer-generated facial images that works effectively for both frontal and angled images.

• This method is based on smoothness property of the faces presented by edges and human skin's characteristic via local entropy.

• Oral Session 2: Watermarking :13:00-15:00

• There are 6 papers presented in this session:

• 1. Fingerprinting for broadcast Content

Distribution System (Japan)

2. Image Watermarking based on Reflectance

Modification (Thailand)

3. Digital Video Watermarking Optimization for

Detecting Replicated two-dimensional

Barcodes (Japan)

• 4. Authentication and Recovery Scheme for

Digital Speech Signal based on DWT (China)

5. Enrichment of Visual Appearance of

Aesthetic QR Code (Japan)

6. Nondestructive Readout of Copyright

Information Embedded in Objects

Fabricated with 3-D printers (Japan)

• The first two papers achieved progressive improvement than related works and the 3rd paper is poorly written, which are not included in the following discussion.

• Two-dimensional barcodes (2D codes) have become common as an authentication way of e-tickets and e-coupons. However, easy replication by using other mobile phone camera is a serious challenge. This paper proposes a method for detecting copied 2D code by a semi-fragile digital video watermark which will be easily destroyed by replication.

• In the fifth paper, the authors considered the characteristic of images to be displayed on a QR code, and proposed an effective method to produce a visually better aesthetic QR code without sacrificing its error correcting capability.

• They also investigated the shape of modules to represent the image to be displayed and considered the trade-off between the visual quality and its readability.

• Actually, we had done a similar work and published our results in IEEE T-MM, 2013.

• The last paper studied a technique to protect the copyrights of digital data for 3-D printers to prevent illegal products from being fabricated with 3-D printers, which has become a serious economic problem.

• The authors proposed a new method for nondestructively reading out the embedded Information, by using thermographic movie files where the binary images of individual frames were summarized to amplify the signals of the cavity patterns and demonstrated the feasibility of automatic readout with 100% accuracy in reading out embedded information.

• To me, this paper presents a new app. of watermarking!

• Oral Session 3: Reversible Data Hiding (RDH)

and Steganalysis :15:30 -17:30

There are 4 RDH papers and 2 Steganalysis papers presented in this session. One of the RDH papers is our contribution to IWDW 2015.

Notice that 3 out of 4 RDH papers (including ours) conducted their data hiding tasks in the encryption domain. We are the only one among the 4 contributions utilizing audios as the host media.

Reversible Data Hiding Papers

• 1. A Commutative Encryption and Reversible Watermarking for Fingerprint Image (India) • 2. Distortion-Free Robust Reversible Watermarking By Modifying and Recording IWT Means of Image Blocks(China) • 3. Reversible Data Hiding for Encrypted Audios by High Order Smoothness (Taiwan) • 4. Completely Separable Reversible Data Hiding in Encrypted Images (China)

• The commutative property , in the title of the first paper, means one can embed and extract watermark in either plain or encrypted fingerprint image, irrespective of order in which encryption or watermarking is applied.

• The encryption and the watermarking keys are shared using secret key sharing mechanism. The biometric database consists of encrypted and watermarked fingerprint images. After successful authentication, watermarking is inverted to get back the plain fingerprint.

• The proposed scheme protects many of the vulnerable points of a biometric authentication system. Also native biometric authentication accuracy remains unaffected due to reversible watermarking. This is the first scheme to include homomorphism between encryption and reversible watermarking. (important!)

• Advantages: 1) higher embedding capacity; 2) better PSNR ratio for the decrypted marked image; 3) use of an optimal key length to preserve security while lowering computational cost; 4) simpler cryptographic key management; 5) smaller side information.

• Authors of the second paper claimed that they have developed a distortion-free robust reversible Image watermarking scheme by using the integer wavelet transform (IWT). They said the proposed method is blind, and provides better performance in the invisibility and the robustness against JPEG/JPEG2000 and noise attacks; however, the audiences found that their embedding capacity is relatively low.

Our contribution to IWDW 2015

• This work presents an RDH method on encrypted audio files, where a data-hider, having no knowledge about the original content, tries to embed some additional data into the encrypted version of the content which was distributed from the content owner.

• A legal receiver, with the pre-negotiated decrypted key, can decrypt the encrypted content and get nearly the same version of the original medium.

• If possible, I will present the work in MM-Security course.

• In the 4th paper, the cover image is first partitioned into non-overlapping blocks and specific encryption is applied to obtain the encrypted image. Then, image difference in the encrypted domain can be calculated based on the homomorphic property of the cryptosystem. The data-hider, who does not know the original image content, may reversibly embed secret data into image difference using a modified version of histogram shifting technique.

• Data extraction is completely separable from (independent of) image decryption, i.e., data extraction can be done either in the encrypted domain or in the decrypted domain, so that it can be applied to different application scenarios.

• In addition, real reversibility is realized, that is, data extraction and image recovery are free of any error.

Two Steganalysis papers

• 1. Video Steganalysis Based on Intra Prediction Mode Calibration (China)

• Some H.264 video steganographic algorithms based on intra prediction mode (IPM) have been proposed. By modulating IPMs over blocks during the intra-coding process, they achieve low computational complexity, considerable capacity and high security. Up to now, there is no existing steganalytic methods found for effective detection.

• The authors of this paper have observed that the existing IPM-based steganography usually modifies IPMs with non-optimal selection rules which violate the encoding principles. By exploiting these weaknesses, they propose a calibration-based video steganalytic scheme, which extracts IPM calibration (IPMC) features to detect IPM-based steganography.

• 2. Feature Selection for High Dimensional Steganalysis (China) • In today′s digital image steganalysis, the dimensionality of

the feature vector is relatively high. This may result in much redundancy and high computational complexity. In the 2nd paper, a feature selection method is proposed to reduce the required dimensionality.

• The main idea is that the element in the extracted feature vector should consistently increase or decrease with the increase of embedding rate for a given steganographic scheme. Experimental results demonstrate that the proposed feature selection method can reduce the dimensionality of the high dimensional feature vector efficiently, and meanwhile the detection accuracy can be well preserved.

Oct. 9 Keynote 2

• Subject: Cyber Forensics – Key Trends and Opportunities (Dr. Vrizlynn Thing, Institute for Infocomm Research, Singapore )

• The borderless cyber space brings challenges to attribution in cyber forensics. Advancement in technology , directly or indirectly , related to anti-forensics, such as big data, data manipulation – tampering - forgery , steganalysis and encryption, and perpetrator’s sophistication and insiders’ knowledge in conceiving (envisioning) novel and trickier fraud (cheating), scam, phishing and information theft, further exacerbate the problem.

• In this talk, Dr. Thing discussed

(i) how the fast paced advancement of

technologies,

(ii) new attack means and techniques,

which are leading to an ever-evolving cyber

threat landscape and challenges in forensics.

To me, this means Digital Forensics is one of interesting and promising new research directions for CS-major students.

• Special Session: Watermarking based on Common Evaluation Criteria : 10:40 -12:40

• 1. Watermarking Method using Concatenated Code for Scaling and Rotation Attacks (Japan) • 2. DCT-OFDM Based Watermarking Scheme Robust against Clipping, Rotation, and Scaling Attacks (Japan) • 3. Robust Imperceptible Video Watermarking for MPEG Compression and DA-AD Conversion Using Visual Masking (Korea) • 4. Detection of frequency-scale modification using robust Audio watermarking based on amplitude modulation (Japan) • 5. Audio Watermarking Using Different Wavelet Filters (Japan)

• The 5 papers presented in special session focused on the development of Robust Watermarking Schemes for Images, Videos and Audios.

• If you are interesting in developing new multimedia Robust watermarking schemes, this session presents a good start point to you..

Short Presentation for Poster (3 min. introduction/summary for each poster)

• Poster Section: 14:25 – 17:30 • 1. Optimal Histogram-pair and Prediction-error Based Reversible Data Hiding for

Medical Images (USA)

• 2. Source Camera Model Identification Using Features from contaminated Sensor Noise (France)

• 3. Blind Watermarking Based on Adaptive Lattice Quantization Index Modulation (China)

• 4. Inter-frame Forgery Detection for Static-background Video Based on MVP Consistency (China)

• 5. Self-embedding Watermarking Scheme Based on MDS Codes (China)

• 6. A New Construction of Tagged Visual Cryptography Scheme (China)

• 7. Synthetic speech detection and audio steganography in VoIP scenarios (Italy)

• 8. An Effective Detection Method Based On Physical Traits of Recaptured Images on LCD Screens (China)

• In this poster session, I spent most of my time with Prof. Yao Zhao (co-author of the last paper, from Beijing Jiaotong University, China), because I think how to effectively detect recaptured images from HD-LCD Screens is a challenging issue we might face, in the near future.

• The detection of recaptured images plays a particular role in public security forensics. Although researches have achieved some progress, low quality of image samples and long time consuming for feature extraction are still prominent problems. (HD Screen + HD Camera = ?)

• From the analysis to the photography process, Dr. Yao Zhao and his co-workers present two effective features for distinguishing high-resolution and high-quality recaptured images from LCD screens.

• One feature is the block effect and blurriness effect caused by JPEG compression, and the other feature is screen effect described by wavelet decomposition with aliasing-enhancement preprocessing.

• Experiments show that the proposed scheme obtains outstanding performances, which is fast and has higher discriminative accuracy.

• IWDW Banquet - Yakatabune Crusing :

18:00 – 20:30 (excellent Views and Wines)

Oral Session 4 : Content Security Oct. 10 09:10 -10:10

• I am very glad to see there is another oral paper from Taiwan; unfortunately, Prof. Umeno (from Kyoto University), invited me to join a breakfast meeting with him, I missed this morning session.

• 1. Authenticated Secret Sharing Scheme Based on General Multi-Exploiting Modification Direction -GMEMD (National Yunlin University of Science & Technology, Taiwan)

• 2. Robust Content-Based Image Hash Functions Using Nested Lattice Codes (Japan)

• 3. An Improved Aspect Ratio Invariant Visual Cryptography Scheme with Flexible Pixel Expansion (Japan)

Oral Session 5 : Forensics (II) 10:40 – 12:00

• There are 3 Forensics related papers presented in the last session of IWDW 2015. Notice that all of them came from China!

• 1. Multiple MP3 Compression Detecting based on the Statistical Properties of Scale Factors (China)

• 2. Detection of Double compression for HEVC videos based on the Co-occurrence Matrix of DCT coefficients (China)

• 3. An Advanced Texture Analysis Scheme for Sharpening Detection (USA, China)

• By revealing the historical traces (ie. # of times) of the audio compression, one could check whether the suspicious MP3 audio has been tampered or not.

• Scale factor is an important parameter in MP3 encoding. In the 1st paper, the variation law of the scale factor with various times of the compression is discovered by analyzing its statistical properties.

• The proposed method can be applied to detect multiple compression operations when the bit rates of the second and third compression are the same.

• Detection of double video compression can reveal partly the video processing history, which is considered to be an important auxiliary means used in video forensics.

• In the 2nd paper, by analyzing the distribution of quantized discrete cosine transform (DCT) coefficients, the authors proposed a scheme to detect double HEVC videos compression under different quantization parameter (QP) values.

• Sharpening is a basic digital image processing manipulation to pursue better image quality worldwide. It is required to detect this kind of operation in image forensics. Huge progress has been made in this area in recent years.

• Overshoot artifact, as an unique phenomenon occurring on image edges after sharpening, is important in sharpening detection

• In the 3rd paper, an advanced method for analyzing overshooting artifact is proposed to boost the detection performance particularly under the slightly sharpening case and overshoot-artifact-control sharpening case. Several experiments has been conducted with the aid of the new scheme to show its best ability in blind sharpening detection, regardless of the strength of overshoot artifact.

Statistical Summary

• In IWDW 2015, there are :

• 2 invited keynote Speeches, all related to Digital Forensics.

• 27 Oral Presentation papers + 8 Poster papers (10 of them are related to digital forensics , 3 of them to Content Security and the others related to Watermarking and Steganalysis. 2 of the 35 papers came from Taiwan, 11 of them came from Japan, and 15 of them came from China)