Abstract
This paper presents a novel multiple description image coding scheme called extended x-tree based multiple description image coding. First, the 2D dual-tree discrete wavelet transform (DDWT) is exerted on the original image to produce 2 wavelet trees. A noise shaping algorithm is used to reduce the redundancies of the resulted wavelet coefficients produced by DDWT, forming the primary parts of two descriptions. In order to enhance the quality of the side reconstruction, a side noise shaping algorithm is then used to produce the supplementary information of side decoding. The primary information from one tree and supplementary information from the other are grouped to form a description. To exploit the strong correlation between two wavelet trees resulted from DDWT, an extended x-tree coding scheme is used to jointly encode two wavelet trees to improve the coding efficiency. Two descriptions are dispatched through separate channels. Since each description includes critical information of the picture, in case that one of the descriptions is lost, the reconstructed image can still keep acceptable fidelity. Simulation results have demonstrated the superior performance and error resilient ability of the proposed algorithm.
Introduction
Transmitting high-quality images over unreliable networks—such as lossy packet-switched networks or unstable wireless connections—presents a significant challenge. Data loss often leads to severe reconstruction artifacts or total failure. Multiple Description Coding (MDC) addresses this by encoding a source into several descriptions sent over independent channels. If all descriptions arrive, a high-quality "central reconstruction" is achieved; if only one arrives, a usable "side reconstruction" is still possible.
Traditional wavelet-based MDC schemes often suffer from the limitations of the standard Discrete Wavelet Transform (DWT), including shift variance and poor directional selectivity. Furthermore, many existing methods fail to fully exploit the correlation between descriptions, leading to suboptimal coding efficiency. This paper proposes a robust MDC scheme leveraging the Dual-Tree Discrete Wavelet Transform (DDWT) and a novel "extended x-tree" joint coding algorithm.
Proposed Methodology
The proposed pipeline utilizes the inherent redundancy of advanced wavelet transforms to enhance error resilience.
Dual-Tree Discrete Wavelet Transform (DDWT)
Unlike standard wavelets, the DDWT produces two parallel sets of wavelet coefficients, forming two distinct trees. While this introduces a 2:1 redundancy, it provides a robust foundation for generating two separate descriptions. The DDWT offers superior directional selectivity, capturing edges at $\pm 15^\circ, \pm 45^\circ,$ and $\pm 75^\circ$, which is critical for preserving complex textures.
Noise Shaping and Description Formation
To manage data overhead, an iterative noise shaping algorithm is employed. This process selectively retains significant coefficients that contribute most to image quality, effectively pruning the trees. The system generates two descriptions: Description 1 contains primary information from Tree A and supplementary information from Tree B, while Description 2 contains primary information from Tree B and supplementary information from Tree A.
Extended X-tree Coding
Because both wavelet trees represent the same underlying image, their structures are highly correlated. The researchers developed an Extended X-tree map to jointly encode both trees. By using a set of nine specific symbols (e.g., SIGROOT, ZSROOT, DESC), the codec can describe the state of both trees simultaneously. This joint encoding drastically reduces the bitrate required for supplementary information compared to independent encoding.
Experimental Results
The algorithm was evaluated using standard benchmark images "Lena" (smooth gradients) and "Barbara" (rich textures). Performance was compared against established MDC techniques (referred to as References [4-6]).
Rate-Distortion Performance
Simulation results demonstrate that the proposed method consistently achieves higher Peak Signal-to-Noise Ratio (PSNR) for a given bitrate. As shown in the rate-distortion curves, the proposed scheme offers a superior tradeoff between central and side reconstruction quality compared to existing literature.
Texture and Error Resilience
The gains were particularly pronounced for the "Barbara" image, where the DDWT’s directional selectivity allowed for better preservation of multi-directional textures at low bitrates. Furthermore, the scheme demonstrated high fidelity in side reconstructions, confirming its robustness against the loss of a complete description.
Conclusion
This paper introduced a novel MDC scheme that effectively utilizes the redundancy of the Dual-Tree Discrete Wavelet Transform. By combining iterative noise shaping with a joint Extended X-tree coding strategy, the proposed method achieves superior coding efficiency and error resilience. Future work will focus on reducing computational complexity for mobile applications and extending the framework to color image sequences.
Summary updated