NO-REFERENCE VIDEO QUALITY ASSESSMENT

Here you can look at some practical examples of the NR-VQA algorithm.
The video sequences are compressed using a H.264 coder with different target average bit-rates. All other settings are kept equal for each sequence. The perceived quality is predicted on a frame-by-frame basis and a final score for the entire sequence is provided as a mean value.

ORIGINAL FRAME (22Mbps) CROP (22Mbps)
Predicted quality = 81
COMPRESSED FRAME (0.8Mbps) CROP (0.8Mbps)
Predicted quality = 62
COMPRESSED FRAME (0.5Mbps) CROP (0.5Mbps)
Predicted quality = 53
 
Frame by frame algorithm output
You can view the entire videos here:

22Mbps

download

0.8Mbps

download

0.5Mbps

download

The algorithm is implemented in MATLAB and source code is available upon request.
CODE