Incorporating Gradient Direction for Assessing Multiple Distortions
Pooryaa Cheraaqee, Azadeh Mansouri, Ahmad Mahmoudi-Aznaveh · 2019
Quantifying the perceived quality of digital images matters when various artifacts cause unpleasant experiences for users of multimedia applications and TV viewers. With an automatic metric for image quality assessment, content providers can monitor what they share and network resources can be assigned in an optimized way. However, it is difficult to model the way humans perceive the quality of an image. Existence of multiple distortions in an image is another challenge, since it makes it more difficult to evaluate the severity of the artifacts in comparison to a situation where there is only one distortion in an image. In this paper, we have adopted the information provided by the direction of gradient vectors, in addition to the gradient magnitudes, so that we can have better performance in dealing with multiple distortions. Doing so, we were able to deliver a full-reference method with an acceptable performance and time complexity for multiply distorted images.