Research Projects of Cagatay DIKICI 

 

1) Dirty Paper Coding with Gaussian Dirty State Informations

In his seminal paper [1], Costa considered the channel coding problem with a squared error distortion constraint on the encoder, where the Gaussian state information S is known to the encoder (See Figure-1). He interpreted the state information known to the encoder as “dirt” (leading to the well-known term of “dirty paper coding” in the data hiding literature) and showed that the capacity in this case is the same as the one where the state information is known by both the encoder and the decoder.

Figure 1: Costa’s “Writing on Dirty Paper” setup.

This problem can be represented as a “blind data hiding” scenario, where the unmarked host S is known by the encoder, but not known by the decoder. The encoder produces the “information-carrying” signal X, where the watermarked signal X + S is operated by the attacker, producing Y.

Figure 2: Proposed “Dirty Paper Coding with Gaussian Dirty State Informations” setup.

In our work (See Figure-2), we consider a channel coding problem where the Gaussian state information S is partially known to the encoder (via Se) and to the decoder (via Sd) (which need not to be same). Since the state information is partially available to the encoder and to the decoder, we call this setup as “Dirty Paper Coding with Gaussian Dirty State Informations”.

Similar to [1], we showed that the capacity in this case is the same as the one where the state information pair (Se,Sd) is known by both the encoder and the decoder.

This problem can be represented as a “privacy-driven data hiding” scenario, where the content owner hires a “data hiding service” from third parties due to the lack of resources or technical know-how, in which the owner does not want to share the unmarked host with the service providers for privacy reasons. Here, S denotes the unmarked host; the watermark encoder (resp. decoder) operates in the absence of S, yet in the potential presence of a sequence Se (resp. Sd) that is correlated with the host. The encoder produces the “information-carrying” signal X, which is subsequently provided to the content owner. Then, the owner carries out data hiding via producing X + S, on which the attacker operates, producing Y. Please refer to [6, 3] for details.

2) Channel Coding under Transmission Rate Constraint

Consider the communication problem between Alice and Bob supplied by a Carrier (See Figure 3). Alice sends a secret message M by inserting it into a host signal S with a power constraint; knowing that Ŝ, a noisy version of the host, is accessible to Bob and to the Carrier at the receiver end. Carrier wants to minimize his communication rate while respecting a quality of services, so compresses the watermarked signal given that Ŝ at the receiver end. Finally, Bob extracts the hidden message M using Ŵ delivered by the Carrier and Ŝ.

Figure 3: A Communication System between Alice and Bob via a Carrier.

Our main contributions can be listed as:

  • The analytic expression of the rate distortion function of the Carrier.
  • The analytic expression of the channel capacity of the overall system.
  • A practical code design for the Gaussian setup is proposed.

The proposed practical coding scheme is based on Trellis Coded Quantization (TCQ) and LDPC codes, where an iterative decoding is performed using belief propagation and BCJR algorithm. For low SNR values, our system operates 1.5 dB away from the channel capacity limit while the embedded message can be decoded with an error rate of Pe ≤ 10−5. Furthermore, our compression scheme operates 0.08 bit per channel use away from the theoretical limits. Please refer to [4] for details.

3) Distributed Video Coding

Within the context of French national project ESSOR (joint work with ENST ParisTech, IRISA Rennes, I3S Nice), we proposed a distributed video coding algorithm. The principal aim is to move the complexity of the encoder to the decoder in order to achieve efficient video coding performance on low power devices.

Figure 4: Wavelet based distributed video coding scheme of ESSOR project.

Distributed source coding is based on fundamental theoretical results of Slepian, Wolf [8] and Wyner and Ziv [9]. In order to apply the distributed coding into video data, the frames of the video are separated in two groups: the key frames (KFs) and the Wyner-Ziv frames (WZs) (See Figure-4). The KFs are encoded with a classical Intra codec such as JPEG2000, and reconstructed by the receiver. At the receiver, the reconstructed KFs also contribute to generate an estimation of the WZFs, called Side Information frames (SIs). These SIs are refined at the receiver using WZ stream, in order to estimate the WZFs.

WZFs are coded using Discrete Wavelet Transform. Afterwards, the quantized coefficients are channel encoded using a capacity achieving low rate channel code such as LDPC. Then, only the syndrome bits are sent through the channel. The key idea of the distributed coding scheme is that the syndrome bits and the SIs are sufficient to reconstruct the WZFs efficiently. Hence, in distributed video coding, the classical motion compensation techniques of the encoder has been replaced by the channel decoding techniques of the receiver. We also proposed an efficient frame interpolation method for generating SIs. Please refer to [5] for details.

4) Attention Based Video Streaming

This work concerns the problem of video streaming from low-power devices where both transmission bandwidth and encoder power are limited. We proposed an attention-based real-time video streaming system that exploits the attentive nature of human viewers for improved video transmission.

Figure 5: Visual field and foveation in a frame.

First, motivated by the fovea-periphery distinction of biological vision systems, the incoming video frame has been partitioned into foveal and periphery regions (See Figure-5) using an application specific attention function. The attention function has been constructed a priori using combinatorial optimization integrated with a back-propagation neural network. Next, a spatial-temporal coding algorithm that exploits the fovea-periphery differentiation is utilized.

The foveal regions are encoded with high spatial resolution while the periphery regions are encoded with a lower spatial resolution. In addition, the periphery regions has also a lower temporal resolution. Finally, the processed video sequence is streamed using a standard streaming server. As an application, we consider the human face video transmission. Our experimental results indicate that even with a very limited amount of training, considerable bandwidth efficiency can be achieved (See Figure-6). Please refer to [2] for details.

Figure 6: Sample frames that are encoded at 25 kbits/sec.

5) Computer Networks

Within the context of French national project DITEMOI [7] (joint work with ENST ParisTech, Thales, Comsis), we are developing robust joint source-channel decoding algorithms for transmitting compressed HTML files through the 802.11n network. 802.11n network layers employ error detection mechanisms such as CRCs or checksums. In this work, these detection mechanisms are exploited in order to correct the errors introduced by the channel. Furthermore, the compression syntax of HTML files is exploited in order to design more efficient source-channel decoding algorithms.

6) Joint Progressive Compression and Watermarking of 3D Meshes

Will be available soon.

7) Compressive Sensing

Will be available soon.

References

[1]
M. Costa. Writing on dirty paper (corresp.). IEEE Trans. Inform. Theory, 29(3):439–441, 1983.
[2]
C. Dikici and H. I. Bozma. Attention Based Video Streaming. Elsevier Signal Processing, Image Communication, vol. 25, no. 10, pp. 745-760, Nov 2010.
[3]
C. Dikici, C. Guillemot, C. Fontaine, K. Idrissi, and A. Baskurt. Dirty Paper Coding with Partial State Information. In IEEE Proc. of International Symposium on Image/Video Communications over fixed and mobile networks (ISIVC), Bilbao, Jul 2008.
[4]
C. Dikici, C. Guillemot, C. Fontaine, K. Idrissi, and A. Baskurt. Joint Data Hiding and Wyner-Ziv Coding, Theory and Practice. EURASIP Journal on Information Security, Feb 2009. (revised).
[5]
C. Dikici, T. Maugey, M. A. Agostini, and O. Crave. Efficient Frame Interpolation for Wyner-Ziv Video Coding. In SPIE VCIP 2009, Jan 2009.
[6]
C. Dikici, M. K. Mihcak, and S. S. Kozat. Gaussian Dirty Paper Coding with Gaussian Dirty State Informations. EURASIP Journal on Advances in Signal Processing, 2010. (submitted).
[7]
DITEMOI. Ditemoi. https://www.research-projects.org/projects/DITEMOI, 2007.
[8]
D. Slepian and J. Wolf. Noiseless coding of correlated information sources. IEEE Trans. Inform. Theory, 19(4):471–480, 1973.
[9]
A. D. Wyner and J. Ziv. The rate-distortion function for source coding with side information at the decoder. IEEE Trans. Inform. Theory, 22(1):1–10, 1976.