英语翻译In this experiment,we want to evaluate the influence of featurecombination to the quality of recovered high-resolution imagewhen using the neighbor embedding algorithm.According to theviewpoint of Freeman et al.(2002),we also attempt to approximatea reasonable weighting factor between matching the localinformation (i.e.the norm luminance) and finding compatible adjacentinformation (i.e.the first-order gradient feature).The distancemeasure between

英语翻译
In this experiment,we want to evaluate the influence of feature
combination to the quality of recovered high-resolution image
when using the neighbor embedding algorithm.According to the
viewpoint of Freeman et al.(2002),we also attempt to approximate
a reasonable weighting factor between matching the local
information (i.e.the norm luminance) and finding compatible adjacent
information (i.e.the first-order gradient feature).The distance
measure between two patches pi and pj is defined as
where DistGrad1 and DistNormL stand for the Euclidean distances of
first-order gradients and normalized luminance between pi and pj,
respectively.a,which is the weighting factor between the two feature
vectors,is approximately set to be 4.The reasoning details are
included in Supplementary material and the setting is verified
empirically as shown in the experiments.
In order to find out the contribution of the feature combination
independently,we don’t apply any other proposed improvements
in the experiments for this part.Instead we adopt the exact procedure
and parameters of the SRNE algorithm which originally employs
first- and second-order gradient concatenation as feature to
do the following three evaluations