The idea of RECO non-reference metric is based on the fact that on the retinal surface natural images are characterized by perfect sharp edges under ideal focusing and in absence of noise, and any edge impairment is perceived as an annoying disturbance, apart some exceptions known as natural factors impairing edge patterns.
Natural factors impairing edge patterns:
- optical distortion
- neuronal thermodynamical/quantum noise
It has been observed that good quality natural images are characterized by high polar edge coherence, whereas edge disruption cause losses of coherence.
These arguments support the conjecture that HVS (Human Vision System) is correlated with polar coherence losses. For more details pls. refer to the paper “THE POLAR EDGE COHERENCE: A QUASI BLIND METRIC FOR VIDEO QUALITY ASSESSMENT“, V. Baroncini et al. According to the paper RECO metric is found to be well-correlated with the subjective video quality (MOS):
Here SROCC means Spearman correlation, this correlation catches a non-linear relationship between elements of data.
I took an open source for computing RECO indexes (https://github.com/aizvorski/video-quality) and modified the code to computing RECO indexes for each frame from a YUV sequence (yuv420p). According to the license I published the modified code – two scripts reco.py and measure_reco.py
Usage of measure_reco.py (reco.py is called as a separated module):
-i input yuv420p file
--width
image width (default 1920)
--height
image height (default 1080)
-n number frames to process, if 0 then the whole stream processed
Example:
python measure_reco.py -i ..\yuv\battlefield_384x320.yuv –width 384 –height 320 -n 3
0 RECO=242.34
1 RECO=194.44
2 RECO=128.96
To check consistency of RECO metrics I take same video sequence encoded with 10Mbps and 20Mbps:
There is still a problem how to map RECO indexes to MOS scores (bad, fair, good, excellent).
NFS Game – 10 Mbps
NFS Game 20 Mbps:
Note: video quality is also display-dependent.
23+ years’ programming and theoretical experience in the computer science fields such as video compression, media streaming and artificial intelligence (co-author of several papers and patents).
the author is looking for new job, my resume