Algorithms of DVC - Distributed Multiple Description Coding - page 136

Due to the exploitation of the temporal correlation between the current frame
and the reference frame at the encoder, residual DVC achieves almost the same
rate-distortion performance as DCT-domain DVC. Meanwhile, the residual DVC
overcomes the difficulty of the combined quantization selection. However, there is
still some spatial correlation remaining in residual frames in [ 18 ]. The rate-distortion
performance should be enhanced further if a proper transform coding with easily
combined quantization methods is adopted to the residual frame. In this section,
a wavelet-domain Wyner-Ziv codec with SPIHT is applied to the residual frame,
which exploits the temporal and spatial correlation at the encoder simultaneously.
Meanwhile, SPIHT brings convenience for the combined quantization of wavelet
coefficients.
On the other hand, the current transform-domain Wyner-Ziv codec assumes the
temporal correlation between the source to be encoded and its side information to
be spatially stationary. In other words, different temporal correlation on the spatial
location is negligible. Though the block-based intra-mode decision in pixel domain
[ 23 , 24 ] is proposed to classify the temporal correlation on spatial location, there is
still much need to exploit intra-mode decision to the transform-domain. In our work,
a mode decision based on temporal and spatial correlation is also proposed. When
the estimated temporal correlation on the spatial location is weak, we use intra-
SPIHT coding on the corresponding wavelet block (WB); otherwise the inter-mode
coder SW-SPIHT is adopted.
Because of the combination of residual coding, SW-SPIHT coding and the intra-
mode decision technique, the proposed DVC is named hybrid DVC (HDVC). The
experimental results show HDVC achieves up to 4.1 dB improvement in PSNR than
pure SW-SPIHT, 3.2 dB improvement than residual DVC, and 3 dB than the latest
DCT-domain DVC of DISCOVER [ 21 ] with some neglectable complexity. Besides,
the intra-mode decision offers up to 2.1 dB improvement at high motion area.

5.4.2 The Proposed Coding Framework

The framework of the proposed HDVC is shown in Fig. 5.13 . Compared to the
current DVC schemes, our contributions mainly focus on the following: (1) wavelet-
domain Wyner-Ziv codec is applied on WBs instead of frames to provide some
conveniences for temporal and spatial correlation analysis; (2) instead of original
frames, the residual frame is used to construct WBs and analyzed to exploit the
temporal and spatial correlation; (3) an intra-mode decision of wavelet domain is
conducted to each WB to exploit different temporal and spatial correlation; (4) the
intra-mode decision strategy based on temporal and spatial criteria is given in [ 25 ].

5.4.2.1 Wavelet-Domain Wyner-Ziv Codec Design for Residual Frames

Residual DVC had been proven to have the same performance as the DCT-domain
DVC due to the exploitation of temporal correlation at the encoder [ 18 ]. Specially,