Efficient HEVC-based video adaptation using transcoding

In a video transmission system, it is important to take into account the great diversity of the network/end-user constraints. On the one hand, video content is typically streamed over a network that is characterized by different bandwidth capacities. In many cases, the bandwidth is insufficient to transfer the video at its original quality. On the other hand, a single video is often played by multiple devices like PCs, laptops, and cell phones. Obviously, a single video would not satisfy their different constraints. These diversities of the network and devices capacity lead to the need for video adaptation techniques, e.g., a reduction of the bit rate or spatial resolution. Video transcoding, which modifies a property of the video without the change of the coding format, has been well-known as an efficient adaptation solution. However, this approach comes along with a high computational complexity, resulting in huge energy consumption in the network and possibly network latency. This presentation provides several optimization strategies for the transcoding process of HEVC (the latest High Efficiency Video Coding standard) video streams. First, the computational complexity of a bit rate transcoder (transrater) is reduced. We proposed several techniques to speed-up the encoder of a transrater, notably a machine-learning-based approach and a novel coding-mode evaluation strategy have been proposed. Moreover, the motion estimation process of the encoder has been optimized with the use of decision theory and the proposed fast search patterns. Second, the issues and challenges of a spatial transcoder have been solved by using machine-learning algorithms. Thanks to their great performance, the proposed techniques are expected to significantly help HEVC gain popularity in a wide range of modern multimedia applications.