VStab-QuAD: A New Video-Stabilization Quality Assessment Database

We introduce VStab-QuAD, a novel Video-Stabilization Quality Assessment Database, consisting of 320 stabilized video sequences obtained from 64 unstable video sequences with different contents (crowd, parallax, running, zooming, etc) using five different strong performing video-stabilization methods. The stabilized videos that are considered contain residual instability-induced image distortions and additional distortions introduced as a side effect of the digital stabilization algorithms applied. In addition, this study presents a comprehensive objective analysis of the video sequences and their validity for video-stabilization quality assessment. Subjective experiments were performed in a controlled environment using the single stimulus continuous quality evaluation (SSCQE) protocol, and their results are included in the database. Therefore, our proposed VStab-QuAD contains (i) the 64 original videos with in-capture distortions due to initial video instability, (ii) the 320 stabilized videos produced with the five different algorithms, and (iii) associated perceptual quality scores.This paper hence contributes VStab-QuAD as a comprehensive public benchmark database designed to facilitate the development of powerful video-stabilization methods and quality assessment metrics in real-life video acquisition scenarios.