Farhad Pakdaman: Complexity reduction and control techniques for power-constrained video coding

Video-based services such as video streaming, E-learning, and social media are now well-integrated into our daily lives, thanks to advanced video encoding techniques which are integrated in many modern consumer devices. However, the high computational complexity and energy consumption of modern video coding standards drain so much power especially from portable devices, rendering such techniques unsuitable. Designing energy-efficient video compression algorithms is crucial for enabling modern video compression in consumer devices.

Consumers’ demand of high-quality video content is challenging for video compression, as video traffic already dominates the Internet. Modern video compression standards such as High Efficiency Video Coding (HEVC) are developed to cope with this demand. However, the compression algorithms used for these standards are very complex and energy consuming. This not only make them unsuitable for power-constrained devices, such as smart phones, but also contributes to high energy consumption in datacenters. Therefore, harnessing the computational complexity in video coding, is essential for enabling modern video coding in various devices, and moving towards energy efficiency. The goal of Farhad Pakdaman’s thesis is to design algorithms and methods that reduce and control the encoding complexity, with a tolerable loss in compression performance.

The contributions of the dissertation can be categorized into two parts.  The first part presents algorithms for reducing the complexity of coding tools, such as motion estimation and intra prediction. Three energy-efficient algorithms are designed that accelerate these complex coding tools in HEVC encoding, while gaining similar compression performance as HEVC.

The second part focuses on scenarios with varying processing power or timing requirements. Examples of such scenarios are smartphones where several applications share the computational resources, or Cloud Gaming where the network delay affects the quality differently in different game states. A machine learning-based scheme is designed that adapts the compression process to the available processing power. This scheme can accurately control the encoding complexity from 100% to 20%, with negligible loss of video quality, beating the current state-of-the-art.

The doctoral dissertation of Farhad Pakdaman in the field of Computing and Electrical Engineering, entitled Complexity Reduction and Control Techniques for Power-Constrained Video Coding, will be publicly examined online at the Faculty of Information Technology and Communication Sciences, Tampere University on Friday, 16 October 2020, at 12 o’clock. The Opponent will be Doctor Bessem Sayadi from Nokia Bell Labs, France. The Custos will be Professor Moncef Gabbouj from Tampere University.

The event can be followed via remote connection.

The dissertation is available online at the  http://urn.fi/URN:ISBN:978-952-03-1707-2