1 Department of Electrical Engineering Stanford University Anne Aaron, Shantanu Rane and Bernd Girod Wyner-Ziv Video Coding with Hash-Based Motion Compensation at the Receiver
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Overview Low-complexity video encoding Wyner-Ziv video codec Hash-based motion-compensation at the receiver Simulation results
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Interframe Video Coding Predictive Interframe Decoder Predictive Interframe Encoder X’ Side Information X
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, “Motion JPEG” Decoder “Motion JPEG” Encoder Video Coding with Low Complexity X’ X Wyner-Ziv Interframe Decoder Wyner-Ziv Intraframe Encoder Side Information [Puri, Ramchandran, Allerton 2002] [Aaron, Zhang, Girod, Asilomar 2002]
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Wyner-Ziv Coding Lossy compression with decoder side information Encoder Decoder Statistically dependent [Wyner and Ziv, 1976] Quantizer Channel Code Channel Decoder Minimum distortion Reconstruction Wyner-Ziv Decoder Wyner-Ziv Encoder “Correlation channel”
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, IDCT W Wyner-Ziv DCT Video Codec Request bits Interpolation/ Extrapolation Reconstruction K Conventional Intraframe coding Conventional Intraframe decoding DCT For each transform band k K’ W’ Y YkYk XkXk Xk’Xk’ level Quantizer DCT Turbo Encoder Buffer Turbo Decoder Extract bit- planes qkqk bit-plane 1 bit-plane 2 bit-plane M k … qk’qk’ Interframe Decoder Intraframe Encoder Side information [Aaron, Rane, Setton, Girod, VCIP 2004] level Quantizer DCT Turbo Encoder Buffer Extract bit- planes Conventional Intraframe coding Conventional Intraframe decoding Interpolation/ Extrapolation DCT Turbo Decoder Reconstruction WZ frames W Key frames K
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Interpolation/ Extrapolation Y Hash-Based Motion Compensation at the Receiver WZ frame W W’ Decoded WZ frames Side information Wyner-Ziv Encoder Wyner-Ziv Decoder “I-P-P-P structure” WZ parity bits Decoded frame/s Motion- compensated Extrapolation Hash Generator (with previous hash store) Hash bits relay motion information to decoder Similar to CRC bits in [Puri, Ramchandran, Allerton 2002] Hash bits Decoded previous frame
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Hash-Based Motion Compensation at the Receiver Hash: quantized subset of DCT coefficients Hash used at turbo decoder as systematic bits Can potentially be used to estimate rate at encoder Encoder: Hash of block may not be sent depending on hash of co-located previous block. Need hash storage. Decoder: Hash used for motion estimation (block-matching). 4x4 DCT
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Compression Results - Salesman Side information generated using previous frame and hash Salesman QCIF sequence at 10fps 7 dB 2 dB
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Compression Results – Hall Monitor 8 dB 2 dB Side information generated using previous frame and hash Hall Monitor QCIF sequence at 10fps
11 Department of Electrical Engineering Stanford University Anne Aaron, Shantanu Rane and Bernd Girod DCT-based Intracoding 247 kbps PSNR Y =33.0 dB Wyner-Ziv DCT codec 256 kbps PSNR Y =39.1 dB GOP=16 Salesman sequence at 10 fps
12 Department of Electrical Engineering Stanford University Anne Aaron, Shantanu Rane and Bernd Girod H.263+ I-P-P-P 249 kbps PSNR Y =43.4 dB GOP=16 Wyner-Ziv DCT codec 256 kbps PSNR Y =39.1 dB GOP=16 Salesman sequence at 10 fps
13 Department of Electrical Engineering Stanford University Anne Aaron, Shantanu Rane and Bernd Girod DCT-based Intracoding 231 kbps PSNR Y =33.3 dB Wyner-Ziv DCT codec 227 kbps PSNR Y =39.1 dB GOP=16 Hall Monitor sequence at 10 fps
14 Department of Electrical Engineering Stanford University Anne Aaron, Shantanu Rane and Bernd Girod H.263+ I-P-P-P 212 kbps PSNR Y =43.0 dB GOP=16 Wyner-Ziv DCT codec 227 kbps PSNR Y =39.1 dB GOP=16 Hall Monitor sequence at 10 fps
Aaron, Rane and Girod: Wyner-Ziv Video Coding October 27, Summary Wyner-Ziv video codec with hash-based motion compensation at the receiver to generate better side information Similar encoding complexity as DCT-based intracoding Compression performance Up to 8 dB better than DCT-based intracoding of frames Performance gap from H.263+ I-P-P coding