Information Technology Reference
In-Depth Information
Detecting blocking artifacts in compressed video. Vlachos proposed an algorithm
based on phase correlation to detect blockiness in MPEG coded SD videos [51]. A
set of sub-sampled images from each frame is cross-correlated to produce a block-
iness measure. The sampling structure chosen is such that each sub-image consists
of a particular pixel from one of the 8
×
8 block used for MPEG compression. Even
though the authors do not test this measure on a database and use an informal subjec-
tive test for one video, Winkler et. al. compare this measure with two other measures
in [52]. However, Winkler et. al. also choose to use a part of the VQEG dataset to
examine performance. Vlachos' measure does not seem to perform too well.
Perceptually significant block-edge impairment metric. Suthaharan proposed the
perceptually significant block-edge impairment metric (PS-BIM) in [53] that uses
luminance masking effects to improve performance. The measure is a ratio of two
terms, each of which is expressed as a linear combination of horizontal and verti-
cal blocking measures. The horizontal and vertical blocking measures are weighted
sums of simple luminance differences where the weights are based on luminance
masking effects. The authors used I-frames from coded sequences to demonstrate
performance. No subjective evaluation was undertaken to evaluate performance.
NR blocking measure for adaptive video processing. Muijs and Kirenko compute
a normalized horizontal gradient D H , norm at a pixel as the ratio of the absolute gra-
dient and the average gradient over a neighboring region in a frame of a video [54].
They then sum D H , norm over the rows to produce a measure S h as a function of the
column. Blocking strength is then a ratio of the mean value of S h at block bound-
aries to the mean value of S h at intermediate positions. A small study was used to
evaluate subjective quality and the algorithm was shown to perform well.
5.2
Multiple Artifact Measurement Based Techniques
No-reference objective quality metric (NROQM). Caviedes and Oberti computed
a set of features including blocking, blurring , and sharpness from the degraded
video in order to assess its quality [55]. Blocking is computed as weighted pixel dif-
ferences between neighboring blocks. Ringing 3 is computed using a combination of
edge detection and 'low-activity' area detection in regions around edges. Clipping -
saturation at low/high pixel values due to numerical precision - is evaluated based
on grey level values of pixels. Noise is measured using a block-based approach. A
histogram-based approach is utilized to compute contrast and a kurtosis based ap-
proach is used for sharpness measurement. A training set of videos (with subjective
MOS ) is used to set parameters in order to combine these measures into a single
score. A separate set of videos was used for testing and high correlation with human
perception was demonstrated.
3
Ringing artifacts are spurious signals that appear around regions of sharp-transitions. In
images, these are seen as 'shimmering' rings around edges.
 
Search WWH ::




Custom Search