Soluções para redução de complexidade da transcodificação de vídeos HEVC para AV1

Abstract

As computer systems and network infrastructures evolve, video technologies evolve together. The current state-of-the-art video coding standard, the High Efficiency Video Coding (HEVC), is focused on the encoding of high-definition (HD) videos. Although it supports ultra-high definition (UHD) videos as well, HEVC tools are not efficient at handling such media. Even so, HEVC is still the standard that achieves the best compression results currently. However, the use of HEVC incurs in non-negligible costs due to royalty policies, especially for video streaming companies. This way, in order to provide an affordable video format for streaming companies, the Alliance for Open Media (AOMedia) was founded. AOMedia developed the AOMedia Video 1 (AV1) format, which is focused on UHD videos for streaming, and launched in early 2018 its first bitstream specification. AV1 is computationally intensive for commercial use, because its encoding time exceeds in 50 times the encoding time of HEVC. Despite of that, it is in the interest of the video streaming industry that the content is offered with no royalty costs to customers. To do so, modifying the format of current content becomes crucial in a business strategy, such as making the cost of the service more accessible or allowing greater compatibility between client and server. To do so, video transcoding techniques are used, in which the encoded file is transformed into another format. This work presents a set of solutions to allow the acceleration of the transcoding process between HEVC and AV1. These solutions are based on the reuse of HEVC bitstream partitioning information to simplify decisions during the AV1 reencoding process. Four proposals have been developed that reach different levels of transcoding time reduction through the use of different aggressiveness levels. A concept called depth level is used to limit the encoder operation in each solution. The most prominent of these four proposals is able to achieve an transcoding time reduction of 35% at an average BD-rate increase of 5.7%. The tests were performed with videos in three different resolutions (HD720, HD1080 and UHD4K) and the best results were achieved precisely in ultra-high resolution videos, the main focus of AV1.