| 摘要 |
<p>Alignment of two video sequences, one being a reference video sequence and the other being a processed video sequence corresponding to the reference video sequence, is performed using selected partial pictures from the reference video sequence. The reference video sequence is evaluated for portions that have maximum information content spatially and/or temporally to select the partial pictures, and the location and size of the partial pictures are stored together with the reference video sequence. The processed video sequence is then initially aligned to the reference video sequence by first finding the locations of features from the partial pictures in the processed video sequence to provide a coarse misalignment error between the two video sequence, and then using a correlation algorithm with a scanning window based on the size of the partial pictures to refine the misalignment error. Once the processed video sequence is aligned with the reference video sequence by using the refined misalignment error, measurement of the differences between the two sequences may be obtained as an indication of the amount of degradation of the processed video sequence with respect to the reference video sequence.</p> |
