A Localized No-Reference Blurriness Measure for Omnidirectional Images and Video

Blurriness is a defect commonly occurring in conventional video but also in omnidirectional video. In this work, we propose a novel no-reference blurriness measure for images captured with omnidirectional video cameras. These images present unique challenges for quality measures due to their size and due to the equirectangular projection which is commonly employed for them. We base upon a state of the art algorithm and adapt it for the specifics of omnidirectional images. Furthermore, we extend it with a coarse-scale blurriness map for measuring spatially varying blur. We present a novel ground truth dataset which was generated by adding spatially varying gaussian blur of different magnitude in a viewport-centric way. Experiments with the proposed algorithm on this dataset show a strong correlation of the localized blurriness measure with the ground truth.