Region-of-interest scrambling for scalable surveillance video using JPEG XR. Paper presented at ACM Multimedia 2009 in Beijing, China. http://portal.acm.org/citation.cfm?id=1631272.1631433

1. Region-of-Interest Scrambling for
Scalable Surveillance Video using JPEG XR
Hosik Sohn, Wesley De Neve, and Yong Man Ro
Image and Video Systems Lab, Department of Electrical Engineering, KAIST, Daejeon, Korea
I. INTRODUCTION
In this paper, we discuss a privacy-protected video surveillance • Contains more transform coefficients than a DC subband, but less
system that makes use of the JPEG XR standard. This standard transform coefficients than an HP subband.
offers a low-complexity solution for the scalable coding of high-
• Random Permutation (RP) was applied to the different transform
resolution images. To address privacy concerns, face regions are
coefficients in the LP subband.
detected and subsequently scrambled in the transform domain,
taking into account the scalability features of JPEG XR. 3) HP and Flexbits Subbands
II. IMAGE CODING USING JPEG XR • Visual effect of scrambled HP
1. Scalable Intra Coding subbands can hardly be seen at
4CIF resolution.
• Low computational complexity, while offering a high image quality
• Even at a low spatial resolution,
and spatial and quality scalability provisions.
face regions with a sufficiently
• Frequency domain in JPEG XR (4 subbands) : DC (1), low pass (15), high resolution cannot be
high pass (240), and Flexbits (256) subband. Fig 3. Visual impact of scrambled HP subbands:
concealed adequately. (a) QCIF resolution and (b) 4CIF resolution.
2. ROI Representation • For this reason, we propose not to scramble HP subbands and
Flexbits subbands.
• Uniform tile layout : each tile has the
same width and height. IV. EXPERIMENTAL RESULTS
• Non-uniform tile layout : tiles may have 1. Visual Results
different widths and heights.
III. SCRAMBLING Fig 1. ROI representation
1. Proposed Encoder Architecture
Secret key
DC Fig 4. Privacy-protected surveillance video: (a) DC, (b) DC + LP, (c) DC + LP + HP, and (d) DC + LP
LBT LBT Q Pred. Scrambling + HP + Flexbits.
• Adaptive
Low pass entropy 2. Bit Stream Overhead Analysis
Adaptive coding
Q Pred. Scrambling Table 1. Bit stream overhead according to the tile size
scan • Fixed
High pass/Flexbits length Tile grid 1x1 MB 5x5 MB 10x10 MB
9 tiles
Adaptive coding Bit rate (Kbit/s) (%) (%) (%)
Q Pred.
scan 629 10.6 771.9 72.2 16.5
955 7.3 482.1 47.6 11.2
Fig 2. Architecture of our modified JPEG XR encoder
1348 4.5 323.0 32.8 7.4
2. Subband-Adaptive Scrambling 1964 2.8 207.9 21.5 4.6
Important factors for scrambling 2809 1.9 135.5 14.2 3.2
4404 1.2 86.8 8.9 1.9
• Visual importance of the subband. 5791 0.5 54.4 5.0 0.6
• Available amount of coded data in the subband. 8158 0.2 35.0 3.0 0.2 Fig. 5. Bit stream overhead introduced by
scrambling
• Level of security offered by the scrambling technique . 3. Security Considerations
• Effect on the coding efficiency. • DC subband in one MB
• Computational complexity of the scrambling technique.  2N+1 combinations
1) Scrambling for DC Subbands (N: the number of bits used to
represent the fixed length part of
Random Sign Inversion (RSI): the DC coefficient)
where D denotes the data to be scrambled and where De denotes the • LP subband in one MB
pseudo-randomly sign-flipped data.  15! Combinations
Random Bit Flipping (RBF) is applied to the DC refinement bits and • Total number of combinations
the level refinement bits: Fig. 6. Average number of bits used to represent
 2N+1 + 15!
the fixed length part of a DC coefficient
B denotes the data to be encrypted while Be denotes the encrypted V. CONCLUSIONS
data. Further, bi denotes the ith bit of B and R denotes the set of
This paper discussed an approach for scrambling privacy-sensitive
pseudo-random bits.
face regions in scalable surveillance video coded using JPEG XR. Our
Each DC coefficient is partitioned into a significant part and a approach is the result of a trade-off between the visual importance of
refinement bits. The significant part is again partitioned into a level subbands, the amount of coded data in the subbands, the level of
value and level refinement bits. security offered by a particular scrambling technique, the effect of
2) Scrambling for LP Subbands scrambling on the coding efficiency, and the computational complexity
of the scrambling technique used. The results show that privacy-
• LP subband : visually less important than a DC subband, but sensitive regions can be successfully concealed with a feasible level of
visually more important than an HP subband. protection.