Quality-Temporal Transcoder Driven by the Jerkiness

Perceptual visual quality metrics: A survey

“…For decoded video without availability of the coding parameters, frame freeze can be simply detected by frame difference [85]; in the cases when the frame rate is available, the jerkiness effect can be evaluated using the frame rate [122,134,192] or more comprehensively, both the frame rate and temporal activity (i.e., motion) [60,91]. The location, number and duration of lost frames were estimated via inter-frame correlation analysis in [109], while lost frames and the density of group dropping were detected by inter-frame dissimilarity to measure fluidity in [126,127], in which it was concluded that, for the same level of frame loss, scattered fluidity breaks introduce less quality degradation than aggregated ones.…”

Temporal Quality Evaluation for Enhancing Compressed Video

“…Key related works on objective temporal quality metric design include [8,9,10,11,12,13]. Feghali et al [8] use frame rate as the scaling factor to adjust Peak-Signal-to-Noise-Ratio (PSNR) and output a spatial-temporal quality score.…”

“…[16,17,10,9], same amount of dropping loss may introduce different temporal quality impact because of different motion activity; higher motion activity usually results in larger temporal quality degradation. Therefore, the dropping severity s m,n should be mapped to s m,n according to motion activity by…”

“…In Refs. [9] and [10], a jerkiness metric based on frame rate and motion activity is proposed. In Ref.…”

Unequal inter-view rate allocation using scalable stereo video coding and an objective stereo video quality measure

“…To this effect, we propose subtracting a jerkiness factor J from the optimistic left PSNR, in such a way to place the resulting measure (Q) closer to the optimistic or pessimistic left PSNR value depending on the motion activity of the video. For the n-th frame, a motion activity measure can be defined by [10] NZ I (1) where N I and N NZ are the number of Intra Macroblocks (MBs) and the number of MBs in which there exist nonzero motion vectors in a local region r of the n-th frame respectively, P I is a weight for Intra MBs that is set to 128 (the maximum of motion vector for a 64×64 motion search window) and mvx i and mvy i are the horizontal and vertical components of the i-th motion vector. We divide a frame spatially into nine regions, which include equal number of MBs, and compute MA r , r=1,…,9, in each region.…”