Control of video processing algorithms based on measured perceptual quality characteristics

摘要

Systems and methods may provide for conducting a real-time perceptual quality analysis of a video, wherein the perceptual quality analysis includes at least one of a noise measurement, a contrast measurement and a sharpness measurement. One or more strength parameters of one or more post-processing modules in a video processing pipe may be set based on the perceptual quality analysis resulting in overall video processing that adapts to the changing perceptual quality of the input. In one example, the strength parameters include at least one of a contrast parameter and a de-noising parameter. The invention results in visually enhanced video at the output of the post-processing module.

主权项

1. A computer implemented method comprising:
receiving an input video; computing perceptual noise in the input video including a smoothed noise estimate from a plurality of noise estimates by sorting the plurality of noise estimates in an M-point window, wherein M is based on one or more of a frame rate of the input video or a predetermined duration of smoothing, and by smoothing the plurality of noise estimates using a Gaussian kernel to obtain the smoothed noise estimate; and controlling de-noising strength in a post-processing pipeline by modifying a standard deviation for a filter based on the smoothed noise estimate.