(a) it is perceptually closest to the original andThe perceptual difference between reproduction and original - PD(r,o) - is then defined by applying band-pass filters to the images and then obtaining their difference using the following formula, which takes into account human contrast sensitivity:
(b) all itís pixels are within the reproduction gamut
Here h is the impulse response of an observation filter, r(x,y)
and o(x,y) are the CIELAB coordinates of the reproduction and original
at the point (x,y) in the image and * represents convolution.
A method for calculating an image which satisfies the two criteria stated
above is given in the paper and this method has the advantage of taking
into account image detail rather than being pixel-based like almost all
the other algorithms reviewed here. However, the accuracy of the difference
metric being minimised is a potential problem with this method, which is
otherwise a very good one.