Relationship modeling of encode quality and encode parameters based on source attributes

A method of predicting bitrate, codec, resolution, or other filter parameters for a filter chain to achieve a full reference video quality score for encoding an input source video, the method comprising:

• identifying a source video quality of the source video;
• identifying a source content complexity of the source video;
• receiving parameter constraints with respect to the parameters;
• applying the source video quality, the source content complexity, and the parameter constraints to a deep neural network (DNN) producing DNN outputs;
• and
• combining the DNN outputs using domain knowledge to provide the filter parameters, as predicted, to the filter chain, such that applying the filter chain to the input source video results in an output video achieving the full reference video quality score.

A method of predicting a full-reference video quality score of a source video after performance of scaling, transcoding, and/or filtering operations, the method comprising:

• identifying a source video quality of the source video;
• identifying a source content complexity of the source video;
• receiving content parameters of the source video;
• receiving player metrics indicative of aspects of playback by a consumer device of an output video corresponding to the source video;
• receiving parameter constraints with respect to parameters of the output video;
• applying the source video quality, the source content complexity, the content parameters, the parameter constraints, and the player metrics to a deep neural network (DNN) producing DNN outputs;
• and
• combining the DNN outputs using domain knowledge to produce an overall predicted quality score of the output video, without accessing the output video.