Improved Video Compression using Perceptual Modeling

摘要

Disclosed are techniques for video stream compression. A motion predicted frame and a current frame of video data are converted from a source color space to a perceptual color space. A delta (Δ) frame is calculated between the current frame and the motion predicted frame. Input color difference values between the current frame and the motion predicted frame are determined and mapped to a normalized output to create a tone mapped color difference image. The tone mapped color difference image and Δ frame are multiplied to obtain a weighted Δ frame. Δ loss is calculated as the difference between the weighted Δ frame and the Δ frame. A variance map of the motion predicted frame is multiplied by the Δ loss to obtain a Δ preservation map. The weighted Δ frame and Δ preservation map are summed to obtain a modified Δ frame which may be transformed, quantized and encoded.

主权项

1. A method comprising:
determining a Δ frame by calculating the difference between each pixel in a current frame and each corresponding pixel in a motion predicted frame, the current frame and the motion predicted frame expressed in a perceptual color space; determining a weighted Δ frame by multiplying a tone mapped color difference image and the Δ frame, the tone mapped color difference image determined by mapping input color difference values between each pixel in the current frame and each corresponding pixel in the motion predicted frame to a normalized output for each pixel; determining a Δ preservation map by multiplying a variance map of the motion predicted frame by the difference between the weighted Δ frame and the Δ frame; summing the weighted Δ frame and Δ preservation map to obtain a modified Δ frame.