Controle Adaptativo de Complexidade do Codificador AV1 baseado em Frente de Pareto e Aprendizado de Máquina.

Abstract

The encoders, responsible for the compression of the digital videos, enable the use of this type of media, as they reduce the data rate, maintaining the image quality. The continuous development and Improvements to digital video coding standards are essential to boost applications based on this type of media, making the market, constantly look for new algorithms and coding standards that reach high compression efficiency. Given this scenario, the AOM (Alliance for Open Media) developed AV1 (AOMedia Video 1), video encoder with a high compression ratio. However, advanced tools and enhancements present in AV1 have a high computational cost, leading to a high coding time. Therefore, the main objective of this work is to contribute with strategies for adaptive complexity control of the AV1 encoder, aiming to optimize the relationship between time and coding efficiency for HD 1080 and UHD 4K resolution videos. For this, a study on the complexity of AV1 coding was carried out, where different points of control were obtained from the analysis of AV1 coding parameters. From this, the Baseline Version of the controller was developed and used to evaluate the viability of developing the adaptive AV1 complexity controller. To improve the dynamic control capacity of the controller, a new version of the controller was developed, called CCAM. In it, machine learning models were used to predict the coding time and classify the input video according to its characteristics. The average error results obtained with the CCAM controller range from 0.11 to 1.88 percentage points at HD 1080 resolution and from 0.14 to 3.33 percentage points at UHD 4K resolution, for a range of encoding time reduction from 10% to 70%. The average BD-Rate values obtained for the CCAM controller in HD 1080 and UHD 4K resolutions are between 2.63% and 54.57% and 2.08% and 38 .30%, respectively. Furthermore, results of a subjective quality assessment indicate a minimal degradation in the subjective quality of the compressed sequences from the CCAM controller developed. It should also be noted that the controllers proposed in this thesis are the first solutions in the literature for the adaptive control of the complexity of the AV1 encoder.