Image quality assessment in first-person videos

First-person videos (FPVs) or egocentric videos provide a huge amount of data for visual lifelogs. The quality assessment of frames in FPVs serves as an important tool, feature or evaluation baseline for not only structuring but also analyzing lifelogs. To develop a frame-quality measure for FPVs, we introduce a new strategy for image quality estimation, called mutual reference (MR), which uses one or more pseudo-reference images to evaluate a test image. We then propose a MR quality estimator, called Local Visual Information (LVI), that primarily measures the relative blur between two images. To apply the MR strategy to FPVs, we propose a mutual reference frame quality assessment for FPVs (MRFQAFPV) framework which incorporates LVI. Our results, using both real and synthetic distortions and objective and subjective tests, demonstrate both methods perform better than existing NR QEs at measuring the quality of frames in FPVs.