Copy the page URI to the clipboard
Mahmood, Asif
(2020).
DOI: https://doi.org/10.21954/ou.ro.000110c8
Abstract
Video coding is the core enabling technology for compressing sequences where inherently very large bandwidth requirements must be reduced to enable efficient transmission and storage. Traditional video codecs (encoder-decoder) generally incur large encoder resources and so are not suitable for resource-scarce devices. Distributed video coding (DVC) has become an attractive alternative for such devices because of its simple encoder structure, which is achieved by shifting the major processing load to the decoder. Despite the established theory, current DVC architectures do not exhibit the same rate-distortion (RD) performance as traditional codecs, a performance gap which is compounded when coding higher resolution sequences. Available DVC codec rate-control options are also inflexible and difficult to guarantee a quality-of- service (QoS) level, while the aggregated encoder-decoder latency at higher spatial resolutions can significantly hinder practical DVC implementations.
This thesis presents a new DVC for Higher Resolution (DVC-HR) framework that addresses the aforementioned limitations arising in encoding higher spatial resolution sequences. It introduces a novel flexible content-aware quantisation (CAQ) mechanism which supports dynamic and robust rate-control to produce maximum output quality by lowering perceptible distortion as well as more efficiently utilising available bandwidth. Crucially, CAQ supports larger transform block sizes which are essential for effective higher resolution DVC. Finally, the latency issue is assuaged with the development of a new dynamic channel-coding algorithm to reduce decoding times thereby affording an important advance for practical DVC realisations.
Rigorous analysis of the DVC-HR framework confirms that it consistently outperforms the community-accepted benchmark DVC codec (DISCOVER) in RD performance, rate-control flexibility, bandwidth utilisation and lower latency at higher spatial resolutions. These contributions represent a notable step towards narrowing the DVC performance gap by enabling more practical, state-of-the-art implementations to better rival traditional codecs.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)
