Abstract: A system for encoding video for film grain synthesis is disclosed. A video encoding system includes a film grain description estimator configured to receive an input video comprising film grain and analyze the input video to generate a film grain description for the input video. The video encoding system includes a video encoder configured to receive the input video and, responsive to a determination that the input video comprises film grain, configure one or more encoder settings based on presence of the film grain to achieve a compression target for the input video. The video encoder is further configured to encode the input video comprising film grain using the encoder settings configured based on the presence of film grain to produce encoded video. The video encoding system is configured to send the encoded video and the film grain description to a video destination.

Abstract: Techniques for automatically configuring settings for a video encoder are described.

Abstract: Techniques for providing content aware uniform quality for quality-defined variable bitrate encoding of videos are described. According to some embodiments, a computer-implemented method includes receiving a request at a content delivery service to encode a live video at a quality level of a plurality of quality levels of a quality-defined variable bitrate encoding, determining features for a frame of the live video, generating a first predicted quantization parameter for the frame by a first machine learning model, trained for a first objective quality metric, receiving an input of the features and the quality level, performing a real time encode of the live video based at least in part on the first predicted quantization parameter to generate an encoded live video, and transmitting the encoded live video to a viewer device.

Abstract: Techniques for selectively preserving film grain for video encoding are described. In certain embodiments, a content delivery system and/or service determines one or more blocks of a frame to preserve film grain in the encoding in contrast to preserving the film grain in the encoding of all of the blocks of the frame.

Abstract: Techniques for training and using machine learning models for video encoding mode selection are described. According to some embodiments, a computer-implemented method includes receiving a live video at a content delivery service, extracting one or more features for a plurality of macroblocks of a frame of the live video, determining an encoding mode from a plurality of encoding modes for each of the plurality of macroblocks of the frame with a machine learning model based at least in part on an input of the one or more features, performing a real time encode of the frame of the live video based at least in part on the determined encoding modes to generate an encoded frame by the content delivery service, and transmitting the encoded frame from the content delivery service to a viewer device.

Abstract: Techniques for providing perceptually motivated video pre-filtering are described.

Abstract: A content streaming system and methodology facilitates the selection of sample adaptive offset filter parameters for use in a sample adaptive filter in encoding content for content streaming. Selection of the sample adaptive offset filter parameters can include multiple active parameter selection modes responsive to block-based statistics. Multiple active parameter selection modes provide increased granularity in the encoded content.

Abstract: Systems and methods are described to process image-based content to reduce the complexity of information used to represent the content, without visible different to a viewing user. Individual images of the content items can be broken into image blocks, each of which can be converted via DCT algorithm to a DCT matrix. Values within the DCT matrix can be compared to just-noticeable-difference (JND) thresholds, below which information is unlikely to be perceptible to viewing users. For each value in the DCT matrix below a corresponding JND threshold, the value can be reduced to zero, thus reducing complexity of the image without perceptible difference to a viewing user and enabling more efficient use of computing resources to store and transmit the image. JND thresholds can be based on one or more models of human perception, including a contrast sensitivity function model, a luminance adjustment model, and a contrast masking model.