Exploração das distorções da projeção ERP para redução de complexidade da codificação de vídeos omnidirecionais

Abstract

The equirectangular projection (ERP) is a pre-processing step that allows omnidirectional videos to be encoded through conventional video coding standards. Althought it enables the encoding of such videos, ERP projection inserts distortions in the video as horizontal stretching in the textures depending on the frame region, and such distortions introduce specific behaviors in video coding tools that do not occur when encoding conventional videos. Therefore, this work explores the distortions casued by equirectangular projection (ERP) to propose a set of complexity reduction techniques for the encoding of omnidirectional video, which are applied to both interframes and intraframe prediction. The complexity reduction techniques applied to interframe prediction are named Early Skip, Reduced SR and Reduced FME. The Early Skip technique is based on performing an early decision to employ the skip prediction mode based on the block homogeneity, the Reduced SR technique consists on reducing the width of the search range during motion estimation, and Reduced FME technique is based on reducing the horizontal precision of fractional motion estimation. When these three techniques are employed together, it is possible to achieve, on average, 22.84% of complexity reduction with small coding efficiency penalties. The complexity reduction techniques designed to accelerate the intraframe prediction are named Reduced Intra Sizes and Reduced Intra Modes. Reduced Intra Sizes technique discards the evaluation of unlikely block sizes and Reduced Intra Modes technique is inherited from another work, and consists on prioritizing the selectional of horizontal intra prediction modes. When both intraprediction techniques are employed together, an average complexity reduction 23.09% is observed whereas the coding efciciently is not significantly changed. All designed techniques – both for interframes and intraframe prediction – exploit the ERP distortion throughout the frame to accelerate the encoding properly. In addition, the designed techniques present results competitive with related works and show that it is advantageous to explore the ERP projection distortions to achieve complexity reduction wishout harming coding efficiency significantly.