Machine Learning for Per-Title Encoding

• DOI: 10.5594/jmi.2022.3154836
• Publication Date: 2022-04-05T19:28:41Z
• Authors: Daniel Silhavy; Christopher Krauss; Anita Chen; Anh-Tu Nguyen; Christoph Muller; Stefan Arbanowski; Stephan Steglich; Louay Bassbouss
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Video streaming content varies in terms of complexity and requires title-specific encoding settings to achieve a certain visual quality. Classic “one-size-fits-all” ladders ignore video-specific characteristics apply the same across all video files. In worst-case scenario, this approach can lead quality impairments, artifacts, or unnecessarily large media A per-title solution has potential significantly decrease storage delivery costs streams while improving perceptual video. Conventional solutions typically require number test encodes, resulting high computational times costs. article, we describe that implements conventional uses its data for machine learning-based improvements. By applying supervised, multivariate regression algorithms like random forest regression, multilayer perceptron (MLP), support vector predict metric (VMAF) values. These values are foundation deriving optimal ladder. As result, encodes eliminated preserving benefits encoding</i> .