Blind Video Inpainting Quality Assessment

To our best knowledge, there are no objective methods for Blind Video Inpainting Quality Assessment BVINQA. Given the cost of subjective quality assessment, it is essential to provide objective assessment methods. Existing methods dedicated to still images could not be used for video images, since the artifacts depending on the movement of objects are not taken into account. Therefore, we propose to develop a BVIN-QA by considering both spatial and temporal aspects of videos and by extracting frame-specific spatial and temporal quality scores. The overall score is provided by using a support vector regression SVR, fed by frame score statistics and trained using subjective MOS scores.The proposed method has been tested on a set of inpainting videos using several kinds of masks and the most known inpainting methods. The obtained results are very promising.