TEWI-Kolloquium: Intra-coding using non-linear prediction, KLT and Texture Synthesis: AV1 encoders open the door to seemingly unconstrained video coding complexity
VeranstaltungsortE.2.42Veranstalter Fakultät für Technische WissenschaftenBeschreibungWe present a comparison between AV1, HEVC and JEM. It reveals that professionally optimized AV1 encoding software is about 32 times as complex as the HEVC encoder software HM. Given the wide attention that AV1 receives and the plans of Google and Facebook to actually use AV1 in commercial environments, the complexity constraints endured during the HEVC development seem to disappear. We present recent results on contour-based prediction for intra coding, a KLT depending on scene and intra prediction direction for coding the prediction error and texture synthesis to replace conventional texture coding for uniform parts of images. Especially contour-based prediction and texture synthesis depend heavily on computer vision algorithms. While typically failures of efficient prediction can be compensated by coding the prediction error and hence a higher data rate, texture synthesis in video coding does not have this safety leash.Vortragende(r)Prof. Dr.-Ing. Jörn Ostermann KontaktKerstin Smounig (kerstin.smounig@aau.at)