Error Resilience in Heterogeneous Visual Communications
Publication or External Link
A critical and challenging aspect of visual communication technologies is to immunize visual information to transmission errors. In order to effectively protect visual content against transmission errors, various kinds of heterogeneities involved in multimedia delivery need to be considered, such as compressed stream characteristics heterogeneity, channel condition heterogeneity, multi-user and multi-hop heterogeneity. The main theme of this dissertation is to explore these heterogeneities involved in error-resilient visual communications to deliver different visual content over heterogeneous networks with good visual quality.
Concurrently transmitting multiple video streams in error-prone environment faces many challenges, such as video content characteristics are heterogeneous, transmission bandwidth is limited, and the user device capabilities vary. These challenges prompt the need for an integrated approach of error protection and resource allocation. One motivation of this dissertation is to develop such an integrated approach for an emerging application of multi-stream video aggregation, i.e. multi-point video conferencing. We propose a distributed multi-point video conferencing system that employs packet division multiplexing access (PDMA)-based error protection and resource allocation, and explore the multi-hop awareness to deliver good and fair visual quality of video streams to end users.
When the transport layer mechanism, such as forward error correction (FEC), cannot provide sufficient error protection on the payload stream, the unrecovered transmission errors may lead to visual distortions at the decoder. In order to mitigate the visual distortions caused by the unrecovered errors, concealment techniques can be applied at the decoder to provide an approximation of the original content. Due to image characteristics heterogeneity, different concealment approaches are necessary to accommodate different nature of the lost image content. We address this heterogeneity issue and propose to apply a classification framework that adaptively selects the suitable error concealment technique for each damaged image area.
The analysis and extensive experimental results in this dissertation demonstrate that the proposed integrated approach of FEC and resource allocation as well as the new classification-based error concealment approach can significantly outperform conventional error-resilient approaches.