SARI 1.1 Performance Chart

SARI 1.1: Performance Charts and Technical Description

Technical Description:

SARI is based on two invariant properties of DCT-based lossy compression. The first property shows that if a DCT coefficient is modified to an integral multiple of a quantization step, which is larger than the steps used in later JPEG compressions, then this coefficient can be exactly reconstructed after later JPEG compression. The second one is the invariant relationships between two coefficients in a block pair before and after JPEG compression. In SARI, we use the second property to generate authentication signature, and use the first property to embed it as watermarks. These properties provide solutions to two major challenges in developing authentication watermarks ( a.k.a. integrity watermarks): how to extract short, invariant, and robust information to substitute fragile hash function, and how to embed information that is guaranteed to survive DCT-based lossy compression to an acceptable extent. Because the first property almost reaches maximum zero-error embedding capacity, in additional to authentication signatures, we also embed the recovery bits for recovering approximate pixel values in corrupted areas. SARI authenticator utilizes the compressed bitstream, and thus avoids rounding errors in reconstructing DCT coefficients. The extraordinary embedding capacity of SARI can be applied to any kinds of information hiding applications.

Tips:

Watermarking is a three-party game: the amount of embedded information, the visual quality degradation, and the robustness of embedded watermark. In SARI 1.1embedder, users can select two kinds of embedded information: authentication bits only, or authentication bits and recovery bits . By dragging the scroll bar, users can select 5 different embedding mode: QRmode = 0 - 4. With embedding watermark at QRmode = 0, users can obtain the best visual quality watermarked image but the least robustness. On the other hand, using QRmode = 4, users get the most robust watermark at the exponse visual quality.

Robustness:

Here is a robustness performance chart shown how the embedded watermark survives JPEG lossy compression using Photoshop 5.0. It is image independent.

	Authentication bits survive	Recovery bits survive
QRmode:0	JPEG Quality >= 9	JPEG Quality >= 9
QRmode:1	JPEG Quality >= 6	JPEG Quality >= 9
QRmode:2	JPEG Quality >= 4	JPEG Quality >= 6
QRmode:3	JPEG Quality >= 3	JPEG Quality >= 6
QRmode:4	JPEG Quality >= 2	JPEG Quality >= 6

This table reads like this: if you select "Auth Only" and "best visual quality" in the embedder then the all the bits of embedded watermark are guaranteed to survive JPEG lossy compression using Photoshop 5.0 Quality large or equal to 9. (Photoshop 5.0 uses a 0-10 scale for compression quality.) This watermarked image can only be authenticated but not recovered. If you select "Auth+Recovery" and "best visual quality", then this watermarked image can be authenticated while it is compressed using Quality>=9 [from first column], and it can be recovered as long as it is compressed using Quality >=9 [from second column].

Visual Quality:

The degradation of visual quality is image dependent. Here is an example of using lenna 512x512 color image after embedding.

PSNR	Auth Only	Auth+Recovery
QRmode:0	48.67 dB	42.60 dB
QRmode:1	46.37 dB	41.89 dB
QRmode:2	44.57 dB	38.01 dB
QRmode:3	43.04 dB	37.63 dB
QRmode:4	39.77 dB	36.42 dB

Embedded Information:

The amout of embedded information is image dependent. Here is an example of using lenna 512x512 color image after using the sharware version of SARI 1.1 embedder.

	Auth Only	Auth+Recovery
QRmode:0	12,288 bits	47,240 bits
QRmode:1	12,288 bits	47,240 bits
QRmode:2	12,288 bits	47,240 bits
QRmode:3	12,288 bits	47,240 bits
QRmode:4	12,288 bits	47,240 bits

These bits all can be exactly extracted from the compressed image with the compressed quality refer to the previous robustness performance chart.

Related Techniques and Application:

Techniques used in SARI 1.1 are partially described in [1,2,3]. Such techniques can be used to implement other features which are currently not demonstrated in SARI shareware. For example,

authentication of motion JPEG and MPEG video,
watermarking for information hiding
find-grain control of trade-off among robustness, watermark length, and visual quality,
public key management functionalities for restricted access control.

References: [ download ]

[1] Ching-Yung Lin and Shih-Fu Chang, "A Robust Image Authentication Method Surviving JPEG Lossy Compression," SPIE Storage and Retrieval of Image/Video Database, EI '98, San Jose, Jan 1998; also in IEEE Trans. on Circuits and Systems for Video Technology, 2000.

[2] Ching-Yung Lin and Shih-Fu Chang, "Issues and Solutions for Authenticating MPEG Video," SPIE Security and Watermarking of Multimedia Contents, EI '99, San Jose, CA, Jan. 1999.

[3] Ching-Yung Lin and Shih-Fu Chang, "Semi-Fragile Watermarking for Authenticating JPEG Visual Content," SPIE Security and Watermarking of Multimedia Contents II, EI '00, San Jose, CA, Jan. 2000.

Contact Us:

SARI includes technologies with patent applications pending.
For technical questions, please contact Ching-Yung Lin or Shih-Fu Chang.
For licensing questions, please contact Robert Schleelein of Columbia Innovation Enterprise (CIE), Ching-Yung Lin or Shih-Fu Chang

Webpage Copyright 2000-2001 by Ching- Yung Lin, ADVENT Lab, Columbia Univeristy
Last updated : Oct 22, 2000

If you are interested in the multimedia authentication topics,
please contact me or sign my guestbook !!