摘要 |
A method of predicting subjective quality ratings of processed video from corresponding human vision model perceptual difference scores obtains perceptual difference scores for a "Worst" quality video training sequence and for a "Best" quality video training sequence. Corresponding subjective quality rating values are assigned to the perceptual difference scores as modified by any single-ended measures of impairments that may exist in the reference video training sequences from which the "Worst" and "Best" quality video training sequences are derived. A conversion function, which may be a piecewise linear function, an "S" curve function or other function that approximates the non-linearities and compression at the extremes of the subjective quality rating scale, is used to produce a conversion curve of calibration values based on the perceptual difference scores for the "Worst" and "Best" quality video training sequences and heuristically derived constants.
|