Reliably combining quality indicators
 
Reliably combining quality indicators 
 
Adriaan Barri, Ann Dooms, Peter Schelkens
 
Abstract 

Objective quality measures based on machinelearning (ML) require fewer computations andare less affected by inaccuracies in the HVSmodels. But they may also yield less transparentquality predictions when the ML responses aredifficult to interpret. The absence ofinterpretability may disguise seriousvulnerabilities in the design of the objectivequality measure.